ViewVC Help

View File | Revision Log | Show Annotations | Download File

/cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.8 by root, Sun May 11 20:08:37 2008 UTC vs.
Revision 1.121 by root, Sat Jun 2 20:13:26 2012 UTC

		1	/*
		2	* libeio implementation
		3	*
		4	* Copyright (c) 2007,2008,2009,2010,2011,2012 Marc Alexander Lehmann <libeio@schmorp.de>
		5	* All rights reserved.
		6	*
		7	* Redistribution and use in source and binary forms, with or without modifica-
		8	* tion, are permitted provided that the following conditions are met:
		9	*
		10	* 1. Redistributions of source code must retain the above copyright notice,
		11	* this list of conditions and the following disclaimer.
		12	*
		13	* 2. Redistributions in binary form must reproduce the above copyright
		14	* notice, this list of conditions and the following disclaimer in the
		15	* documentation and/or other materials provided with the distribution.
		16	*
		17	* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
		18	* WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
		19	* CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
		20	* EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
		21	* CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
		22	* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
		23	* OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
		24	* WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
		25	* ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
		26	* OF THE POSSIBILITY OF SUCH DAMAGE.
		27	*
		28	* Alternatively, the contents of this file may be used under the terms of
		29	* the GNU General Public License ("GPL") version 2 or any later version,
		30	* in which case the provisions of the GPL are applicable instead of
		31	* the above. If you wish to allow the use of your version of this file
		32	* only under the terms of the GPL and not to allow others to use your
		33	* version of this file under the BSD license, indicate your decision
		34	* by deleting the provisions above and replace them with the notice
		35	* and other provisions required by the GPL. If you do not delete the
		36	* provisions above, a recipient may use your version of this file under
		37	* either the BSD or the GPL.
		38	*/
		39
		40	#ifndef _WIN32
		41	# include "config.h"
		42	#endif
		43
1	#include "eio.h"	44	#include "eio.h"
		45	#include "ecb.h"
		46
		47	#ifdef EIO_STACKSIZE
		48	# define X_STACKSIZE EIO_STACKSIZE
		49	#endif
2	#include "xthread.h"	50	#include "xthread.h"
3		51
4	#include <errno.h>	52	#include <errno.h>
5	#include <stddef.h>	53	#include <stddef.h>
6	#include <stdlib.h>	54	#include <stdlib.h>
…		…
10	#include <sys/stat.h>	58	#include <sys/stat.h>
11	#include <limits.h>	59	#include <limits.h>
12	#include <fcntl.h>	60	#include <fcntl.h>
13	#include <assert.h>	61	#include <assert.h>
14		62
		63	/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
		64	/* intptr_t only comes from stdint.h, says idiot openbsd coder */
		65	#if HAVE_STDINT_H
		66	# include <stdint.h>
		67	#endif
		68
		69	#ifndef ECANCELED
		70	# define ECANCELED EDOM
		71	#endif
		72	#ifndef ELOOP
		73	# define ELOOP EDOM
		74	#endif
		75
		76	#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
		77	# define ENOTSOCK WSAENOTSOCK
		78	#endif
		79
		80	static void eio_destroy (eio_req *req);
		81
15	#ifndef EIO_FINISH	82	#ifndef EIO_FINISH
16	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0	83	# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
17	#endif	84	#endif
18		85
19	#ifndef EIO_DESTROY	86	#ifndef EIO_DESTROY
…		…
22		89
23	#ifndef EIO_FEED	90	#ifndef EIO_FEED
24	# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)	91	# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
25	#endif	92	#endif
26		93
		94	#ifndef EIO_FD_TO_WIN32_HANDLE
		95	# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
		96	#endif
		97	#ifndef EIO_WIN32_HANDLE_TO_FD
		98	# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
		99	#endif
		100
		101	#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
		102
		103	#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
		104
27	#ifdef _WIN32	105	#ifdef _WIN32
28		106
29	/*doh*/	107	#undef PAGESIZE
		108	#define PAGESIZE 4096 /* GetSystemInfo? */
		109
		110	/* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
		111
		112	#ifdef EIO_STRUCT_STATI64
		113	/* look at perl's non-sloppy stat */
		114	#define stat(path,buf) _stati64 (path,buf)
		115	#define fstat(fd,buf) _fstati64 (fd,buf)
		116	#endif
		117	#define lstat(path,buf) stat (path,buf)
		118	#define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
		119	#define mkdir(path,mode) _mkdir (path)
		120	#define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
		121
		122	#define chmod(path,mode) _chmod (path, mode)
		123	#define dup(fd) _dup (fd)
		124	#define dup2(fd1,fd2) _dup2 (fd1, fd2)
		125
		126	#define fchmod(fd,mode) EIO_ENOSYS ()
		127	#define chown(path,uid,gid) EIO_ENOSYS ()
		128	#define fchown(fd,uid,gid) EIO_ENOSYS ()
		129	#define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
		130	#define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
		131	#define mknod(path,mode,dev) EIO_ENOSYS ()
		132	#define sync() EIO_ENOSYS ()
		133	#define readlink(path,buf,s) EIO_ENOSYS ()
		134	#define statvfs(path,buf) EIO_ENOSYS ()
		135	#define fstatvfs(fd,buf) EIO_ENOSYS ()
		136
		137	/* rename() uses MoveFile, which fails to overwrite */
		138	#define rename(old,neu) eio__rename (old, neu)
		139
		140	static int
		141	eio__rename (const char *old, const char *neu)
		142	{
		143	if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
		144	return 0;
		145
		146	/* should steal _dosmaperr */
		147	switch (GetLastError ())
		148	{
		149	case ERROR_FILE_NOT_FOUND:
		150	case ERROR_PATH_NOT_FOUND:
		151	case ERROR_INVALID_DRIVE:
		152	case ERROR_NO_MORE_FILES:
		153	case ERROR_BAD_NETPATH:
		154	case ERROR_BAD_NET_NAME:
		155	case ERROR_BAD_PATHNAME:
		156	case ERROR_FILENAME_EXCED_RANGE:
		157	errno = ENOENT;
		158	break;
		159
		160	default:
		161	errno = EACCES;
		162	break;
		163	}
		164
		165	return -1;
		166	}
		167
		168	/* we could even stat and see if it exists */
		169	static int
		170	symlink (const char *old, const char *neu)
		171	{
		172	#if WINVER >= 0x0600
		173	if (CreateSymbolicLink (neu, old, 1))
		174	return 0;
		175
		176	if (CreateSymbolicLink (neu, old, 0))
		177	return 0;
		178	#endif
		179
		180	return EIO_ERRNO (ENOENT, -1);
		181	}
		182
		183	/* POSIX API only */
		184	#define CreateHardLink(neu,old,flags) 0
		185	#define CreateSymbolicLink(neu,old,flags) 0
		186
		187	struct statvfs
		188	{
		189	int dummy;
		190	};
		191
		192	#define DT_DIR EIO_DT_DIR
		193	#define DT_REG EIO_DT_REG
		194	#define D_NAME(entp) entp.cFileName
		195	#define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
30		196
31	#else	197	#else
32		198
33	# include "config.h"
34	# include <sys/time.h>	199	#include <sys/time.h>
35	# include <sys/select.h>	200	#include <sys/select.h>
		201	#include <sys/statvfs.h>
36	# include <unistd.h>	202	#include <unistd.h>
		203	#include <signal.h>
		204	#include <dirent.h>
		205
		206	#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
		207	#include <sys/mman.h>
		208	#endif
		209
		210	#define D_NAME(entp) entp->d_name
		211
		212	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
		213	#if __FreeBSD__ || __NetBSD__ || __OpenBSD__
		214	#define _DIRENT_HAVE_D_TYPE /* sigh */
		215	#define D_INO(de) (de)->d_fileno
		216	#define D_NAMLEN(de) (de)->d_namlen
		217	#elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
		218	#define D_INO(de) (de)->d_ino
		219	#endif
		220
		221	#ifdef _D_EXACT_NAMLEN
		222	#undef D_NAMLEN
		223	#define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
		224	#endif
		225
		226	#ifdef _DIRENT_HAVE_D_TYPE
		227	#define D_TYPE(de) (de)->d_type
		228	#endif
		229
		230	#ifndef EIO_STRUCT_DIRENT
		231	#define EIO_STRUCT_DIRENT struct dirent
		232	#endif
		233
		234	#endif
		235
		236	#if HAVE_UTIMES
37	# include <utime.h>	237	# include <utime.h>
38	# include <signal.h>
39	# include <dirent.h>
40
41	# ifndef EIO_STRUCT_DIRENT
42	# define EIO_STRUCT_DIRENT struct dirent
43	# endif	238	#endif
44		239
45	#endif	240	#if HAVE_SYS_SYSCALL_H
46		241	# include <sys/syscall.h>
47	# ifndef EIO_STRUCT_STAT
48	# define EIO_STRUCT_STAT struct stat
49	# endif	242	#endif
		243
		244	#if HAVE_SYS_PRCTL_H
		245	# include <sys/prctl.h>
		246	#endif
50		247
51	#if HAVE_SENDFILE	248	#if HAVE_SENDFILE
52	# if __linux	249	# if __linux
53	# include <sys/sendfile.h>	250	# include <sys/sendfile.h>
54	# elif __freebsd	251	# elif __FreeBSD__ || defined __APPLE__
55	# include <sys/socket.h>	252	# include <sys/socket.h>
56	# include <sys/uio.h>	253	# include <sys/uio.h>
57	# elif __hpux	254	# elif __hpux
58	# include <sys/socket.h>	255	# include <sys/socket.h>
59	# elif __solaris /* not yet */	256	# elif __solaris
60	# include <sys/sendfile.h>	257	# include <sys/sendfile.h>
61	# else	258	# else
62	# error sendfile support requested but not available	259	# error sendfile support requested but not available
63	# endif	260	# endif
64	#endif	261	#endif
65		262
66	/* number of seconds after which an idle threads exit */	263	#ifndef D_TYPE
67	#define IDLE_TIMEOUT 10	264	# define D_TYPE(de) 0
		265	#endif
		266	#ifndef D_INO
		267	# define D_INO(de) 0
		268	#endif
		269	#ifndef D_NAMLEN
		270	# define D_NAMLEN(entp) strlen (D_NAME (entp))
		271	#endif
68		272
69	/* used for struct dirent, AIX doesn't provide it */	273	/* used for struct dirent, AIX doesn't provide it */
70	#ifndef NAME_MAX	274	#ifndef NAME_MAX
71	# define NAME_MAX 4096	275	# define NAME_MAX 4096
72	#endif	276	#endif
73		277
		278	/* used for readlink etc. */
		279	#ifndef PATH_MAX
		280	# define PATH_MAX 4096
		281	#endif
		282
74	/* buffer size for various temporary buffers */	283	/* buffer size for various temporary buffers */
75	#define EIO_BUFSIZE 65536	284	#define EIO_BUFSIZE 65536
76		285
77	#define dBUF \	286	#define dBUF \
78	char *eio_buf; \
79	X_LOCK (wrklock); \
80	self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \	287	char *eio_buf = malloc (EIO_BUFSIZE); \
81	X_UNLOCK (wrklock); \	288	errno = ENOMEM; \
82	if (!eio_buf) \	289	if (!eio_buf) \
83	return -1;	290	return -1
		291
		292	#define FUBd \
		293	free (eio_buf)
84		294
85	#define EIO_TICKS ((1000000 + 1023) >> 10)	295	#define EIO_TICKS ((1000000 + 1023) >> 10)
		296
		297	/*****************************************************************************/
		298
		299	struct tmpbuf
		300	{
		301	void *ptr;
		302	int len;
		303	};
		304
		305	static void *
		306	tmpbuf_get (struct tmpbuf *buf, int len)
		307	{
		308	if (buf->len < len)
		309	{
		310	free (buf->ptr);
		311	buf->ptr = malloc (buf->len = len);
		312	}
		313
		314	return buf->ptr;
		315	}
		316
		317	struct tmpbuf;
		318
		319	#if _POSIX_VERSION >= 200809L
		320	#define HAVE_AT 1
		321	#define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
		322	#ifndef O_SEARCH
		323	#define O_SEARCH O_RDONLY
		324	#endif
		325	#else
		326	#define HAVE_AT 0
		327	static const char *wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path);
		328	#endif
		329
		330	struct eio_pwd
		331	{
		332	#if HAVE_AT
		333	int fd;
		334	#endif
		335	int len;
		336	char str[1]; /* actually, a 0-terminated canonical path */
		337	};
		338
		339	/*****************************************************************************/
		340
		341	#define ETP_PRI_MIN EIO_PRI_MIN
		342	#define ETP_PRI_MAX EIO_PRI_MAX
		343
		344	struct etp_worker;
		345
		346	#define ETP_REQ eio_req
		347	#define ETP_DESTROY(req) eio_destroy (req)
		348	static int eio_finish (eio_req *req);
		349	#define ETP_FINISH(req) eio_finish (req)
		350	static void eio_execute (struct etp_worker *self, eio_req *req);
		351	#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
		352
		353	/*****************************************************************************/
		354
		355	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
		356
		357	/* calculate time difference in ~1/EIO_TICKS of a second */
		358	ecb_inline int
		359	tvdiff (struct timeval *tv1, struct timeval *tv2)
		360	{
		361	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
		362	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
		363	}
		364
		365	static unsigned int started, idle, wanted = 4;
86		366
87	static void (*want_poll_cb) (void);	367	static void (*want_poll_cb) (void);
88	static void (*done_poll_cb) (void);	368	static void (*done_poll_cb) (void);
89		369
90	static unsigned int max_poll_time = 0;	370	static unsigned int max_poll_time; /* reslock */
91	static unsigned int max_poll_reqs = 0;	371	static unsigned int max_poll_reqs; /* reslock */
92		372
93	/* calculcate time difference in ~1/EIO_TICKS of a second */	373	static unsigned int nreqs; /* reqlock */
94	static int tvdiff (struct timeval *tv1, struct timeval *tv2)	374	static unsigned int nready; /* reqlock */
95	{	375	static unsigned int npending; /* reqlock */
96	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	376	static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
97	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	377	static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
98	}
99		378
100	static unsigned int started, idle, wanted;	379	static xmutex_t wrklock;
		380	static xmutex_t reslock;
		381	static xmutex_t reqlock;
		382	static xcond_t reqwait;
		383
		384	typedef struct etp_worker
		385	{
		386	struct tmpbuf tmpbuf;
		387
		388	/* locked by wrklock */
		389	struct etp_worker *prev, *next;
		390
		391	xthread_t tid;
		392
		393	#ifdef ETP_WORKER_COMMON
		394	ETP_WORKER_COMMON
		395	#endif
		396	} etp_worker;
		397
		398	static etp_worker wrk_first; /* NOT etp */
		399
		400	#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
		401	#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
101		402
102	/* worker threads management */	403	/* worker threads management */
103	static mutex_t wrklock = X_MUTEX_INIT;
104		404
105	typedef struct worker {	405	static void
106	/* locked by wrklock */
107	struct worker *prev, *next;
108
109	thread_t tid;
110
111	/* locked by reslock, reqlock or wrklock */
112	eio_req *req; /* currently processed request */
113	void *dbuf;
114	DIR *dirp;
115	} worker;
116
117	static worker wrk_first = { &wrk_first, &wrk_first, 0 };
118
119	static void worker_clear (worker *wrk)	406	etp_worker_clear (etp_worker *wrk)
120	{	407	{
121	if (wrk->dirp)	408	}
122	{
123	closedir (wrk->dirp);
124	wrk->dirp = 0;
125	}
126		409
127	if (wrk->dbuf)	410	static void ecb_cold
128	{	411	etp_worker_free (etp_worker *wrk)
		412	{
129	free (wrk->dbuf);	413	free (wrk->tmpbuf.ptr);
130	wrk->dbuf = 0;
131	}
132	}
133		414
134	static void worker_free (worker *wrk)
135	{
136	wrk->next->prev = wrk->prev;	415	wrk->next->prev = wrk->prev;
137	wrk->prev->next = wrk->next;	416	wrk->prev->next = wrk->next;
138		417
139	free (wrk);	418	free (wrk);
140	}	419	}
141		420
142	static volatile unsigned int nreqs, nready, npending;	421	static unsigned int
143	static volatile unsigned int max_idle = 4;	422	etp_nreqs (void)
144
145	static mutex_t reslock = X_MUTEX_INIT;
146	static mutex_t reqlock = X_MUTEX_INIT;
147	static cond_t reqwait = X_COND_INIT;
148
149	unsigned int eio_nreqs (void)
150	{	423	{
		424	int retval;
		425	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		426	retval = nreqs;
		427	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
151	return nreqs;	428	return retval;
152	}	429	}
153		430
154	unsigned int eio_nready (void)	431	static unsigned int
		432	etp_nready (void)
155	{	433	{
156	unsigned int retval;	434	unsigned int retval;
157		435
158	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	436	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
159	retval = nready;	437	retval = nready;
160	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	438	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
161		439
162	return retval;	440	return retval;
163	}	441	}
164		442
165	unsigned int eio_npending (void)	443	static unsigned int
		444	etp_npending (void)
166	{	445	{
167	unsigned int retval;	446	unsigned int retval;
168		447
169	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	448	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
170	retval = npending;	449	retval = npending;
171	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);	450	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
172		451
173	return retval;	452	return retval;
174	}	453	}
175		454
176	unsigned int eio_nthreads (void)	455	static unsigned int
		456	etp_nthreads (void)
177	{	457	{
178	unsigned int retval;	458	unsigned int retval;
179		459
180	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);	460	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
181	retval = started;	461	retval = started;
…		…
188	* a somewhat faster data structure might be nice, but	468	* a somewhat faster data structure might be nice, but
189	* with 8 priorities this actually needs <20 insns	469	* with 8 priorities this actually needs <20 insns
190	* per shift, the most expensive operation.	470	* per shift, the most expensive operation.
191	*/	471	*/
192	typedef struct {	472	typedef struct {
193	eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */	473	ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
194	int size;	474	int size;
195	} reqq;	475	} etp_reqq;
196		476
197	static reqq req_queue;	477	static etp_reqq req_queue;
198	static reqq res_queue;	478	static etp_reqq res_queue;
199		479
200	static int reqq_push (reqq *q, eio_req *req)	480	static void ecb_noinline ecb_cold
		481	reqq_init (etp_reqq *q)
		482	{
		483	int pri;
		484
		485	for (pri = 0; pri < ETP_NUM_PRI; ++pri)
		486	q->qs[pri] = q->qe[pri] = 0;
		487
		488	q->size = 0;
		489	}
		490
		491	static int ecb_noinline
		492	reqq_push (etp_reqq *q, ETP_REQ *req)
201	{	493	{
202	int pri = req->pri;	494	int pri = req->pri;
203	req->next = 0;	495	req->next = 0;
204		496
205	if (q->qe[pri])	497	if (q->qe[pri])
…		…
211	q->qe[pri] = q->qs[pri] = req;	503	q->qe[pri] = q->qs[pri] = req;
212		504
213	return q->size++;	505	return q->size++;
214	}	506	}
215		507
216	static eio_req *reqq_shift (reqq *q)	508	static ETP_REQ * ecb_noinline
		509	reqq_shift (etp_reqq *q)
217	{	510	{
218	int pri;	511	int pri;
219		512
220	if (!q->size)	513	if (!q->size)
221	return 0;	514	return 0;
222		515
223	--q->size;	516	--q->size;
224		517
225	for (pri = EIO_NUM_PRI; pri--; )	518	for (pri = ETP_NUM_PRI; pri--; )
226	{	519	{
227	eio_req *req = q->qs[pri];	520	eio_req *req = q->qs[pri];
228		521
229	if (req)	522	if (req)
230	{	523	{
…		…
236	}	529	}
237		530
238	abort ();	531	abort ();
239	}	532	}
240		533
241	static void grp_try_feed (eio_req *grp)	534	static int ecb_cold
		535	etp_init (void (*want_poll)(void), void (*done_poll)(void))
242	{	536	{
243	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	537	X_MUTEX_CREATE (wrklock);
244	{	538	X_MUTEX_CREATE (reslock);
245	int old_len = grp->size;	539	X_MUTEX_CREATE (reqlock);
		540	X_COND_CREATE (reqwait);
246		541
247	EIO_FEED (grp);	542	reqq_init (&req_queue);
		543	reqq_init (&res_queue);
248		544
249	/* stop if no progress has been made */	545	wrk_first.next =
250	if (old_len == grp->size)	546	wrk_first.prev = &wrk_first;
251	{
252	grp->feed = 0;
253	break;
254	}
255	}
256	}
257		547
258	static int eio_finish (eio_req *req);	548	started = 0;
		549	idle = 0;
		550	nreqs = 0;
		551	nready = 0;
		552	npending = 0;
259		553
260	static int grp_dec (eio_req *grp)	554	want_poll_cb = want_poll;
261	{	555	done_poll_cb = done_poll;
262	--grp->size;
263		556
264	/* call feeder, if applicable */
265	grp_try_feed (grp);
266
267	/* finish, if done */
268	if (!grp->size && grp->int1)
269	return eio_finish (grp);
270	else
271	return 0;	557	return 0;
272	}	558	}
273		559
274	void eio_destroy (eio_req *req)
275	{
276	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
277	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
278
279	EIO_DESTROY (req);
280	}
281
282	static int eio_finish (eio_req *req)
283	{
284	int res = EIO_FINISH (req);
285
286	if (req->grp)
287	{
288	int res2;
289	eio_req *grp = req->grp;
290
291	/* unlink request */
292	if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
293	if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
294
295	if (grp->grp_first == req)
296	grp->grp_first = req->grp_next;
297
298	res2 = grp_dec (grp);
299
300	if (!res && res2)
301	res = res2;
302	}
303
304	eio_destroy (req);
305
306	return res;
307	}
308
309	void eio_grp_cancel (eio_req *grp)
310	{
311	for (grp = grp->grp_first; grp; grp = grp->grp_next)
312	eio_cancel (grp);
313	}
314
315	void eio_cancel (eio_req *req)
316	{
317	X_LOCK (wrklock);
318	req->flags |= EIO_FLAG_CANCELLED;
319	X_UNLOCK (wrklock);
320
321	eio_grp_cancel (req);
322	}
323
324	X_THREAD_PROC (eio_proc);	560	X_THREAD_PROC (etp_proc);
325		561
326	static void start_thread (void)	562	static void ecb_cold
		563	etp_start_thread (void)
327	{	564	{
328	worker *wrk = calloc (1, sizeof (worker));	565	etp_worker *wrk = calloc (1, sizeof (etp_worker));
329		566
330	/*TODO*/	567	/*TODO*/
331	assert (("unable to allocate worker thread data", !wrk));	568	assert (("unable to allocate worker thread data", wrk));
332		569
333	X_LOCK (wrklock);	570	X_LOCK (wrklock);
334		571
335	if (thread_create (&wrk->tid, eio_proc, (void *)wrk))	572	if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
336	{	573	{
337	wrk->prev = &wrk_first;	574	wrk->prev = &wrk_first;
338	wrk->next = wrk_first.next;	575	wrk->next = wrk_first.next;
339	wrk_first.next->prev = wrk;	576	wrk_first.next->prev = wrk;
340	wrk_first.next = wrk;	577	wrk_first.next = wrk;
…		…
344	free (wrk);	581	free (wrk);
345		582
346	X_UNLOCK (wrklock);	583	X_UNLOCK (wrklock);
347	}	584	}
348		585
		586	static void
349	static void maybe_start_thread (void)	587	etp_maybe_start_thread (void)
350	{	588	{
351	if (eio_nthreads () >= wanted)	589	if (ecb_expect_true (etp_nthreads () >= wanted))
352	return;	590	return;
353		591
354	/* todo: maybe use idle here, but might be less exact */	592	/* todo: maybe use idle here, but might be less exact */
355	if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ())	593	if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
356	return;	594	return;
357		595
358	start_thread ();	596	etp_start_thread ();
359	}	597	}
360		598
361	void eio_submit (eio_req *req)	599	static void ecb_cold
		600	etp_end_thread (void)
362	{	601	{
363	++nreqs;
364
365	X_LOCK (reqlock);
366	++nready;
367	reqq_push (&req_queue, req);
368	X_COND_SIGNAL (reqwait);
369	X_UNLOCK (reqlock);
370
371	maybe_start_thread ();
372	}
373
374	static void end_thread (void)
375	{
376	eio_req *req = calloc (1, sizeof (eio_req));	602	eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
377		603
378	req->type = EIO_QUIT;	604	req->type = -1;
379	req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;	605	req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
380		606
381	X_LOCK (reqlock);	607	X_LOCK (reqlock);
382	reqq_push (&req_queue, req);	608	reqq_push (&req_queue, req);
383	X_COND_SIGNAL (reqwait);	609	X_COND_SIGNAL (reqwait);
384	X_UNLOCK (reqlock);	610	X_UNLOCK (reqlock);
…		…
386	X_LOCK (wrklock);	612	X_LOCK (wrklock);
387	--started;	613	--started;
388	X_UNLOCK (wrklock);	614	X_UNLOCK (wrklock);
389	}	615	}
390		616
391	void eio_set_max_poll_time (double nseconds)	617	static int
392	{
393	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
394	max_poll_time = nseconds;
395	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
396	}
397
398	void eio_set_max_poll_reqs (unsigned int maxreqs)
399	{
400	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
401	max_poll_reqs = maxreqs;
402	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
403	}
404
405	void eio_set_max_idle (unsigned int nthreads)
406	{
407	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
408	max_idle = nthreads <= 0 ? 1 : nthreads;
409	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
410	}
411
412	void eio_set_min_parallel (unsigned int nthreads)
413	{
414	if (wanted < nthreads)
415	wanted = nthreads;
416	}
417
418	void eio_set_max_parallel (unsigned int nthreads)
419	{
420	if (wanted > nthreads)
421	wanted = nthreads;
422
423	while (started > wanted)
424	end_thread ();
425	}
426
427	int eio_poll (void)	618	etp_poll (void)
428	{	619	{
429	int maxreqs = max_poll_reqs;	620	unsigned int maxreqs;
		621	unsigned int maxtime;
430	struct timeval tv_start, tv_now;	622	struct timeval tv_start, tv_now;
431	eio_req *req;
432		623
		624	X_LOCK (reslock);
		625	maxreqs = max_poll_reqs;
		626	maxtime = max_poll_time;
		627	X_UNLOCK (reslock);
		628
433	if (max_poll_time)	629	if (maxtime)
434	gettimeofday (&tv_start, 0);	630	gettimeofday (&tv_start, 0);
435		631
436	for (;;)	632	for (;;)
437	{	633	{
		634	ETP_REQ *req;
		635
438	maybe_start_thread ();	636	etp_maybe_start_thread ();
439		637
440	X_LOCK (reslock);	638	X_LOCK (reslock);
441	req = reqq_shift (&res_queue);	639	req = reqq_shift (&res_queue);
442		640
443	if (req)	641	if (req)
…		…
451	X_UNLOCK (reslock);	649	X_UNLOCK (reslock);
452		650
453	if (!req)	651	if (!req)
454	return 0;	652	return 0;
455		653
		654	X_LOCK (reqlock);
456	--nreqs;	655	--nreqs;
		656	X_UNLOCK (reqlock);
457		657
458	if (req->type == EIO_GROUP && req->size)	658	if (ecb_expect_false (req->type == EIO_GROUP && req->size))
459	{	659	{
460	req->int1 = 1; /* mark request as delayed */	660	req->int1 = 1; /* mark request as delayed */
461	continue;	661	continue;
462	}	662	}
463	else	663	else
464	{	664	{
465	int res = eio_finish (req);	665	int res = ETP_FINISH (req);
466	if (res)	666	if (ecb_expect_false (res))
467	return res;	667	return res;
468	}	668	}
469		669
470	if (maxreqs && !--maxreqs)	670	if (ecb_expect_false (maxreqs && !--maxreqs))
471	break;	671	break;
472		672
473	if (max_poll_time)	673	if (maxtime)
474	{	674	{
475	gettimeofday (&tv_now, 0);	675	gettimeofday (&tv_now, 0);
476		676
477	if (tvdiff (&tv_start, &tv_now) >= max_poll_time)	677	if (tvdiff (&tv_start, &tv_now) >= maxtime)
478	break;	678	break;
479	}	679	}
480	}	680	}
481		681
482	errno = EAGAIN;	682	errno = EAGAIN;
483	return -1;	683	return -1;
484	}	684	}
485		685
		686	static void
		687	etp_cancel (ETP_REQ *req)
		688	{
		689	req->cancelled = 1;
		690
		691	eio_grp_cancel (req);
		692	}
		693
		694	static void
		695	etp_submit (ETP_REQ *req)
		696	{
		697	req->pri -= ETP_PRI_MIN;
		698
		699	if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
		700	if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
		701
		702	if (ecb_expect_false (req->type == EIO_GROUP))
		703	{
		704	/* I hope this is worth it :/ */
		705	X_LOCK (reqlock);
		706	++nreqs;
		707	X_UNLOCK (reqlock);
		708
		709	X_LOCK (reslock);
		710
		711	++npending;
		712
		713	if (!reqq_push (&res_queue, req) && want_poll_cb)
		714	want_poll_cb ();
		715
		716	X_UNLOCK (reslock);
		717	}
		718	else
		719	{
		720	X_LOCK (reqlock);
		721	++nreqs;
		722	++nready;
		723	reqq_push (&req_queue, req);
		724	X_COND_SIGNAL (reqwait);
		725	X_UNLOCK (reqlock);
		726
		727	etp_maybe_start_thread ();
		728	}
		729	}
		730
		731	static void ecb_cold
		732	etp_set_max_poll_time (double nseconds)
		733	{
		734	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		735	max_poll_time = nseconds * EIO_TICKS;
		736	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		737	}
		738
		739	static void ecb_cold
		740	etp_set_max_poll_reqs (unsigned int maxreqs)
		741	{
		742	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
		743	max_poll_reqs = maxreqs;
		744	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
		745	}
		746
		747	static void ecb_cold
		748	etp_set_max_idle (unsigned int nthreads)
		749	{
		750	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		751	max_idle = nthreads;
		752	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		753	}
		754
		755	static void ecb_cold
		756	etp_set_idle_timeout (unsigned int seconds)
		757	{
		758	if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
		759	idle_timeout = seconds;
		760	if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
		761	}
		762
		763	static void ecb_cold
		764	etp_set_min_parallel (unsigned int nthreads)
		765	{
		766	if (wanted < nthreads)
		767	wanted = nthreads;
		768	}
		769
		770	static void ecb_cold
		771	etp_set_max_parallel (unsigned int nthreads)
		772	{
		773	if (wanted > nthreads)
		774	wanted = nthreads;
		775
		776	while (started > wanted)
		777	etp_end_thread ();
		778	}
		779
		780	/*****************************************************************************/
		781
		782	static void
		783	grp_try_feed (eio_req *grp)
		784	{
		785	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
		786	{
		787	grp->flags &= ~EIO_FLAG_GROUPADD;
		788
		789	EIO_FEED (grp);
		790
		791	/* stop if no progress has been made */
		792	if (!(grp->flags & EIO_FLAG_GROUPADD))
		793	{
		794	grp->feed = 0;
		795	break;
		796	}
		797	}
		798	}
		799
		800	static int
		801	grp_dec (eio_req *grp)
		802	{
		803	--grp->size;
		804
		805	/* call feeder, if applicable */
		806	grp_try_feed (grp);
		807
		808	/* finish, if done */
		809	if (!grp->size && grp->int1)
		810	return eio_finish (grp);
		811	else
		812	return 0;
		813	}
		814
		815	static void
		816	eio_destroy (eio_req *req)
		817	{
		818	if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
		819	if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
		820
		821	EIO_DESTROY (req);
		822	}
		823
		824	static int
		825	eio_finish (eio_req *req)
		826	{
		827	int res = EIO_FINISH (req);
		828
		829	if (req->grp)
		830	{
		831	int res2;
		832	eio_req *grp = req->grp;
		833
		834	/* unlink request */
		835	if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
		836	if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
		837
		838	if (grp->grp_first == req)
		839	grp->grp_first = req->grp_next;
		840
		841	res2 = grp_dec (grp);
		842
		843	if (!res)
		844	res = res2;
		845	}
		846
		847	eio_destroy (req);
		848
		849	return res;
		850	}
		851
		852	void
		853	eio_grp_cancel (eio_req *grp)
		854	{
		855	for (grp = grp->grp_first; grp; grp = grp->grp_next)
		856	eio_cancel (grp);
		857	}
		858
		859	void
		860	eio_cancel (eio_req *req)
		861	{
		862	etp_cancel (req);
		863	}
		864
		865	void
		866	eio_submit (eio_req *req)
		867	{
		868	etp_submit (req);
		869	}
		870
		871	unsigned int
		872	eio_nreqs (void)
		873	{
		874	return etp_nreqs ();
		875	}
		876
		877	unsigned int
		878	eio_nready (void)
		879	{
		880	return etp_nready ();
		881	}
		882
		883	unsigned int
		884	eio_npending (void)
		885	{
		886	return etp_npending ();
		887	}
		888
		889	unsigned int ecb_cold
		890	eio_nthreads (void)
		891	{
		892	return etp_nthreads ();
		893	}
		894
		895	void ecb_cold
		896	eio_set_max_poll_time (double nseconds)
		897	{
		898	etp_set_max_poll_time (nseconds);
		899	}
		900
		901	void ecb_cold
		902	eio_set_max_poll_reqs (unsigned int maxreqs)
		903	{
		904	etp_set_max_poll_reqs (maxreqs);
		905	}
		906
		907	void ecb_cold
		908	eio_set_max_idle (unsigned int nthreads)
		909	{
		910	etp_set_max_idle (nthreads);
		911	}
		912
		913	void ecb_cold
		914	eio_set_idle_timeout (unsigned int seconds)
		915	{
		916	etp_set_idle_timeout (seconds);
		917	}
		918
		919	void ecb_cold
		920	eio_set_min_parallel (unsigned int nthreads)
		921	{
		922	etp_set_min_parallel (nthreads);
		923	}
		924
		925	void ecb_cold
		926	eio_set_max_parallel (unsigned int nthreads)
		927	{
		928	etp_set_max_parallel (nthreads);
		929	}
		930
		931	int eio_poll (void)
		932	{
		933	return etp_poll ();
		934	}
		935
486	/*****************************************************************************/	936	/*****************************************************************************/
487	/* work around various missing functions */	937	/* work around various missing functions */
488		938
489	#if !HAVE_PREADWRITE
490	# define pread aio_pread
491	# define pwrite aio_pwrite
492
493	/*
494	* make our pread/pwrite safe against themselves, but not against
495	* normal read/write by using a mutex. slows down execution a lot,
496	* but that's your problem, not mine.
497	*/
498	static mutex_t preadwritelock = X_MUTEX_INIT;
499
500	static ssize_t pread (int fd, void *buf, size_t count, off_t offset)
501	{
502	ssize_t res;
503	off_t ooffset;
504
505	X_LOCK (preadwritelock);
506	ooffset = lseek (fd, 0, SEEK_CUR);
507	lseek (fd, offset, SEEK_SET);
508	res = read (fd, buf, count);
509	lseek (fd, ooffset, SEEK_SET);
510	X_UNLOCK (preadwritelock);
511
512	return res;
513	}
514
515	static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset)
516	{
517	ssize_t res;
518	off_t ooffset;
519
520	X_LOCK (preadwritelock);
521	ooffset = lseek (fd, 0, SEEK_CUR);
522	lseek (fd, offset, SEEK_SET);
523	res = write (fd, buf, count);
524	lseek (fd, offset, SEEK_SET);
525	X_UNLOCK (preadwritelock);
526
527	return res;
528	}
529	#endif
530
531	#ifndef HAVE_FUTIMES	939	#ifndef HAVE_UTIMES
532		940
		941	# undef utimes
533	# define utimes(path,times) aio_utimes (path, times)	942	# define utimes(path,times) eio__utimes (path, times)
534	# define futimes(fd,times) aio_futimes (fd, times)
535		943
		944	static int
536	static int aio_utimes (const char *filename, const struct timeval times[2])	945	eio__utimes (const char *filename, const struct timeval times[2])
537	{	946	{
538	if (times)	947	if (times)
539	{	948	{
540	struct utimbuf buf;	949	struct utimbuf buf;
541		950
…		…
546	}	955	}
547	else	956	else
548	return utime (filename, 0);	957	return utime (filename, 0);
549	}	958	}
550		959
		960	#endif
		961
		962	#ifndef HAVE_FUTIMES
		963
		964	# undef futimes
		965	# define futimes(fd,times) eio__futimes (fd, times)
		966
		967	static int
551	static int aio_futimes (int fd, const struct timeval tv[2])	968	eio__futimes (int fd, const struct timeval tv[2])
552	{	969	{
553	errno = ENOSYS;	970	errno = ENOSYS;
554	return -1;	971	return -1;
555	}	972	}
556		973
557	#endif	974	#endif
558		975
559	#if !HAVE_FDATASYNC	976	#if !HAVE_FDATASYNC
		977	# undef fdatasync
560	# define fdatasync fsync	978	# define fdatasync(fd) fsync (fd)
561	#endif	979	#endif
		980
		981	static int
		982	eio__syncfs (int fd)
		983	{
		984	int res;
		985
		986	#if HAVE_SYS_SYNCFS
		987	res = (int)syscall (__NR_syncfs, (int)(fd));
		988	#else
		989	res = EIO_ENOSYS ();
		990	#endif
		991
		992	if (res < 0 && errno == ENOSYS && fd >= 0)
		993	sync ();
		994
		995	return res;
		996	}
		997
		998	/* sync_file_range always needs emulation */
		999	static int
		1000	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
		1001	{
		1002	#if HAVE_SYNC_FILE_RANGE
		1003	int res;
		1004
		1005	if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
		1006	|| EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
		1007	|| EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
		1008	{
		1009	flags = 0
		1010	| (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
		1011	| (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
		1012	| (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
		1013	}
		1014
		1015	res = sync_file_range (fd, offset, nbytes, flags);
		1016
		1017	if (!res || errno != ENOSYS)
		1018	return res;
		1019	#endif
		1020
		1021	/* even though we could play tricks with the flags, it's better to always
		1022	* call fdatasync, as that matches the expectation of its users best */
		1023	return fdatasync (fd);
		1024	}
		1025
		1026	static int
		1027	eio__fallocate (int fd, int mode, off_t offset, size_t len)
		1028	{
		1029	#if HAVE_LINUX_FALLOCATE
		1030	return fallocate (fd, mode, offset, len);
		1031	#else
		1032	return EIO_ENOSYS ();
		1033	#endif
		1034	}
562		1035
563	#if !HAVE_READAHEAD	1036	#if !HAVE_READAHEAD
		1037	# undef readahead
564	# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self)	1038	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
565		1039
		1040	static eio_ssize_t
566	static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self)	1041	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
567	{	1042	{
568	size_t todo = count;	1043	size_t todo = count;
569	dBUF;	1044	dBUF;
570		1045
571	while (todo > 0)	1046	while (todo > 0)
…		…
575	pread (fd, eio_buf, len, offset);	1050	pread (fd, eio_buf, len, offset);
576	offset += len;	1051	offset += len;
577	todo -= len;	1052	todo -= len;
578	}	1053	}
579		1054
580	errno = 0;	1055	FUBd;
		1056
		1057	/* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
		1058	/* but not for e.g. EIO or eof, so we also never fail */
581	return count;	1059	return 0;
582	}	1060	}
583		1061
584	#endif
585
586	#if !HAVE_READDIR_R
587	# define readdir_r aio_readdir_r
588
589	static mutex_t readdirlock = X_MUTEX_INIT;
590
591	static int readdir_r (DIR *dirp, EIO_STRUCT_DIRENT *ent, EIO_STRUCT_DIRENT **res)
592	{
593	EIO_STRUCT_DIRENT *e;
594	int errorno;
595
596	X_LOCK (readdirlock);
597
598	e = readdir (dirp);
599	errorno = errno;
600
601	if (e)
602	{
603	*res = ent;
604	strcpy (ent->d_name, e->d_name);
605	}
606	else
607	*res = 0;
608
609	X_UNLOCK (readdirlock);
610
611	errno = errorno;
612	return e ? 0 : -1;
613	}
614	#endif	1062	#endif
615		1063
616	/* sendfile always needs emulation */	1064	/* sendfile always needs emulation */
		1065	static eio_ssize_t
617	static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self)	1066	eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
618	{	1067	{
		1068	eio_ssize_t written = 0;
619	ssize_t res;	1069	eio_ssize_t res;
620		1070
621	if (!count)	1071	if (!count)
622	return 0;	1072	return 0;
623		1073
		1074	for (;;)
		1075	{
		1076	#ifdef __APPLE__
		1077	# undef HAVE_SENDFILE /* broken, as everything on os x */
		1078	#endif
624	#if HAVE_SENDFILE	1079	#if HAVE_SENDFILE
625	# if __linux	1080	# if __linux
		1081	off_t soffset = offset;
626	res = sendfile (ofd, ifd, &offset, count);	1082	res = sendfile (ofd, ifd, &soffset, count);
627		1083
628	# elif __freebsd	1084	# elif __FreeBSD__
629	/*	1085	/*
630	* Of course, the freebsd sendfile is a dire hack with no thoughts	1086	* Of course, the freebsd sendfile is a dire hack with no thoughts
631	* wasted on making it similar to other I/O functions.	1087	* wasted on making it similar to other I/O functions.
632	*/	1088	*/
633	{
634	off_t sbytes;	1089	off_t sbytes;
635	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	1090	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
636		1091
637	if (res < 0 && sbytes)	1092	#if 0 /* according to the manpage, this is correct, but broken behaviour */
638	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */	1093	/* freebsd' sendfile will return 0 on success */
		1094	/* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
		1095	/* not on e.g. EIO or EPIPE - sounds broken */
		1096	if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
639	res = sbytes;	1097	res = sbytes;
640	}	1098	#endif
		1099
		1100	/* according to source inspection, this is correct, and useful behaviour */
		1101	if (sbytes)
		1102	res = sbytes;
		1103
		1104	# elif defined __APPLE__
		1105	off_t sbytes = count;
		1106	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
		1107
		1108	/* according to the manpage, sbytes is always valid */
		1109	if (sbytes)
		1110	res = sbytes;
641		1111
642	# elif __hpux	1112	# elif __hpux
643	res = sendfile (ofd, ifd, offset, count, 0, 0);	1113	res = sendfile (ofd, ifd, offset, count, 0, 0);
644		1114
645	# elif __solaris	1115	# elif __solaris
646	{
647	struct sendfilevec vec;	1116	struct sendfilevec vec;
648	size_t sbytes;	1117	size_t sbytes;
649		1118
650	vec.sfv_fd = ifd;	1119	vec.sfv_fd = ifd;
651	vec.sfv_flag = 0;	1120	vec.sfv_flag = 0;
652	vec.sfv_off = offset;	1121	vec.sfv_off = offset;
653	vec.sfv_len = count;	1122	vec.sfv_len = count;
654		1123
655	res = sendfilev (ofd, &vec, 1, &sbytes);	1124	res = sendfilev (ofd, &vec, 1, &sbytes);
656		1125
657	if (res < 0 && sbytes)	1126	if (res < 0 && sbytes)
658	res = sbytes;	1127	res = sbytes;
659	}
660		1128
661	# endif	1129	# endif
		1130
		1131	#elif defined (_WIN32) && 0
		1132	/* does not work, just for documentation of what would need to be done */
		1133	/* actually, cannot be done like this, as TransmitFile changes the file offset, */
		1134	/* libeio guarantees that the file offset does not change, and windows */
		1135	/* has no way to get an independent handle to the same file description */
		1136	HANDLE h = TO_SOCKET (ifd);
		1137	SetFilePointer (h, offset, 0, FILE_BEGIN);
		1138	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
		1139
662	#else	1140	#else
663	res = -1;	1141	res = EIO_ENOSYS ();
664	errno = ENOSYS;
665	#endif	1142	#endif
666		1143
		1144	/* we assume sendfile can copy at least 128mb in one go */
		1145	if (res <= 128 * 1024 * 1024)
		1146	{
		1147	if (res > 0)
		1148	written += res;
		1149
		1150	if (written)
		1151	return written;
		1152
		1153	break;
		1154	}
		1155	else
		1156	{
		1157	/* if we requested more, then probably the kernel was lazy */
		1158	written += res;
		1159	offset += res;
		1160	count -= res;
		1161
		1162	if (!count)
		1163	return written;
		1164	}
		1165	}
		1166
667	if (res < 0	1167	if (res < 0
668	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK	1168	&& (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
		1169	/* BSDs */
		1170	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
		1171	|| errno == ENOTSUP
		1172	#endif
		1173	#ifdef EOPNOTSUPP /* windows */
		1174	|| errno == EOPNOTSUPP /* BSDs */
		1175	#endif
669	#if __solaris	1176	#if __solaris
670	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE	1177	|| errno == EAFNOSUPPORT || errno == EPROTOTYPE
671	#endif	1178	#endif
672	)	1179	)
673	)	1180	)
…		…
677		1184
678	res = 0;	1185	res = 0;
679		1186
680	while (count)	1187	while (count)
681	{	1188	{
682	ssize_t cnt;	1189	eio_ssize_t cnt;
683		1190
684	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);	1191	cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
685		1192
686	if (cnt <= 0)	1193	if (cnt <= 0)
687	{	1194	{
…		…
699		1206
700	offset += cnt;	1207	offset += cnt;
701	res += cnt;	1208	res += cnt;
702	count -= cnt;	1209	count -= cnt;
703	}	1210	}
		1211
		1212	FUBd;
704	}	1213	}
705		1214
706	return res;	1215	return res;
707	}	1216	}
708		1217
709	/* read a full directory */	1218	#ifdef PAGESIZE
710	static void scandir_ (eio_req *req, worker *self)	1219	# define eio_pagesize() PAGESIZE
		1220	#else
		1221	static intptr_t
		1222	eio_pagesize (void)
711	{	1223	{
712	DIR *dirp;	1224	static intptr_t page;
713	union
714	{
715	EIO_STRUCT_DIRENT d;
716	char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
717	} *u;
718	EIO_STRUCT_DIRENT *entp;
719	char *name, *names;
720	int memlen = 4096;
721	int memofs = 0;
722	int res = 0;
723		1225
724	X_LOCK (wrklock);	1226	if (!page)
725	self->dirp = dirp = opendir (req->ptr1);	1227	page = sysconf (_SC_PAGESIZE);
726	self->dbuf = u = malloc (sizeof (*u));
727	req->flags |= EIO_FLAG_PTR2_FREE;
728	req->ptr2 = names = malloc (memlen);
729	X_UNLOCK (wrklock);
730		1228
731	if (dirp && u && names)	1229	return page;
732	for (;;)	1230	}
		1231	#endif
		1232
		1233	static void
		1234	eio_page_align (void **addr, size_t *length)
		1235	{
		1236	intptr_t mask = eio_pagesize () - 1;
		1237
		1238	/* round down addr */
		1239	intptr_t adj = mask & (intptr_t)*addr;
		1240
		1241	*addr = (void *)((intptr_t)*addr - adj);
		1242	*length += adj;
		1243
		1244	/* round up length */
		1245	*length = (*length + mask) & ~mask;
		1246	}
		1247
		1248	#if !_POSIX_MEMLOCK
		1249	# define eio__mlockall(a) EIO_ENOSYS ()
		1250	#else
		1251
		1252	static int
		1253	eio__mlockall (int flags)
		1254	{
		1255	#if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
		1256	extern int mallopt (int, int);
		1257	mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
		1258	#endif
		1259
		1260	if (EIO_MCL_CURRENT != MCL_CURRENT
		1261	|| EIO_MCL_FUTURE != MCL_FUTURE)
		1262	{
		1263	flags = 0
		1264	| (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
		1265	| (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
		1266	}
		1267
		1268	return mlockall (flags);
		1269	}
		1270	#endif
		1271
		1272	#if !_POSIX_MEMLOCK_RANGE
		1273	# define eio__mlock(a,b) EIO_ENOSYS ()
		1274	#else
		1275
		1276	static int
		1277	eio__mlock (void *addr, size_t length)
		1278	{
		1279	eio_page_align (&addr, &length);
		1280
		1281	return mlock (addr, length);
		1282	}
		1283
		1284	#endif
		1285
		1286	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
		1287	# define eio__msync(a,b,c) EIO_ENOSYS ()
		1288	#else
		1289
		1290	static int
		1291	eio__msync (void *mem, size_t len, int flags)
		1292	{
		1293	eio_page_align (&mem, &len);
		1294
		1295	if (EIO_MS_ASYNC != MS_SYNC
		1296	|| EIO_MS_INVALIDATE != MS_INVALIDATE
		1297	|| EIO_MS_SYNC != MS_SYNC)
		1298	{
		1299	flags = 0
		1300	| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1301	| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1302	| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1303	}
		1304
		1305	return msync (mem, len, flags);
		1306	}
		1307
		1308	#endif
		1309
		1310	static int
		1311	eio__mtouch (eio_req *req)
		1312	{
		1313	void *mem = req->ptr2;
		1314	size_t len = req->size;
		1315	int flags = req->int1;
		1316
		1317	eio_page_align (&mem, &len);
		1318
		1319	{
		1320	intptr_t addr = (intptr_t)mem;
		1321	intptr_t end = addr + len;
		1322	intptr_t page = eio_pagesize ();
		1323
		1324	if (addr < end)
		1325	if (flags & EIO_MT_MODIFY) /* modify */
		1326	do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1327	else
		1328	do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
		1329	}
		1330
		1331	return 0;
		1332	}
		1333
		1334	/*****************************************************************************/
		1335	/* requests implemented outside eio_execute, because they are so large */
		1336
		1337	static void
		1338	eio__lseek (eio_req *req)
		1339	{
		1340	/* this usually gets optimised away completely, or your compiler sucks, */
		1341	/* or the whence constants really are not 0, 1, 2 */
		1342	int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
		1343	: req->int2 == EIO_SEEK_CUR ? SEEK_CUR
		1344	: req->int2 == EIO_SEEK_END ? SEEK_END
		1345	: req->int2;
		1346
		1347	req->offs = lseek (req->int1, req->offs, whence);
		1348	req->result = req->offs == (off_t)-1 ? -1 : 0;
		1349	}
		1350
		1351	/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
		1352	static int
		1353	eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
		1354	{
		1355	const char *rel = path;
		1356	char *res;
		1357	char *tmp1, *tmp2;
		1358	#if SYMLOOP_MAX > 32
		1359	int symlinks = SYMLOOP_MAX;
		1360	#else
		1361	int symlinks = 32;
		1362	#endif
		1363
		1364	errno = EINVAL;
		1365	if (!rel)
		1366	return -1;
		1367
		1368	errno = ENOENT;
		1369	if (!*rel)
		1370	return -1;
		1371
		1372	res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
		1373	tmp1 = res + PATH_MAX;
		1374	tmp2 = tmp1 + PATH_MAX;
		1375
		1376	#if 0 /* disabled, the musl way to do things is just too racy */
		1377	#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
		1378	/* on linux we may be able to ask the kernel */
		1379	{
		1380	int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
		1381
		1382	if (fd >= 0)
733	{	1383	{
		1384	sprintf (tmp1, "/proc/self/fd/%d", fd);
		1385	req->result = readlink (tmp1, res, PATH_MAX);
		1386	close (fd);
		1387
		1388	/* here we should probably stat the open file and the disk file, to make sure they still match */
		1389
		1390	if (req->result > 0)
		1391	goto done;
		1392	}
		1393	else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
		1394	return;
		1395	}
		1396	#endif
		1397	#endif
		1398
		1399	if (*rel != '/')
		1400	{
		1401	int len;
		1402
734	errno = 0;	1403	errno = ENOENT;
735	readdir_r (dirp, &u->d, &entp);	1404	if (wd == EIO_INVALID_WD)
		1405	return -1;
		1406
		1407	if (wd == EIO_CWD)
		1408	{
		1409	if (!getcwd (res, PATH_MAX))
		1410	return -1;
736		1411
737	if (!entp)	1412	len = strlen (res);
		1413	}
		1414	else
		1415	memcpy (res, wd->str, len = wd->len);
		1416
		1417	if (res [1]) /* only use if not / */
		1418	res += len;
		1419	}
		1420
		1421	while (*rel)
		1422	{
		1423	eio_ssize_t len, linklen;
		1424	const char *beg = rel;
		1425
		1426	while (*rel && *rel != '/')
		1427	++rel;
		1428
		1429	len = rel - beg;
		1430
		1431	if (!len) /* skip slashes */
		1432	{
738	break;	1433	++rel;
		1434	continue;
		1435	}
739		1436
740	name = entp->d_name;	1437	if (beg [0] == '.')
		1438	{
		1439	if (len == 1)
		1440	continue; /* . - nop */
741		1441
742	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))	1442	if (beg [1] == '.' && len == 2)
		1443	{
		1444	/* .. - back up one component, if possible */
		1445
		1446	while (res != tmpbuf->ptr)
		1447	if (*--res == '/')
		1448	break;
		1449
		1450	continue;
		1451	}
		1452	}
		1453
		1454	errno = ENAMETOOLONG;
		1455	if (res + 1 + len + 1 >= tmp1)
		1456	return -1;
		1457
		1458	/* copy one component */
		1459	*res = '/';
		1460	memcpy (res + 1, beg, len);
		1461
		1462	/* zero-terminate, for readlink */
		1463	res [len + 1] = 0;
		1464
		1465	/* now check if it's a symlink */
		1466	linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
		1467
		1468	if (linklen < 0)
743	{	1469	{
		1470	if (errno != EINVAL)
		1471	return -1;
		1472
		1473	/* it's a normal directory. hopefully */
		1474	res += len + 1;
		1475	}
		1476	else
		1477	{
		1478	/* yay, it was a symlink - build new path in tmp2 */
744	int len = strlen (name) + 1;	1479	int rellen = strlen (rel);
745		1480
		1481	errno = ENAMETOOLONG;
		1482	if (linklen + 1 + rellen >= PATH_MAX)
		1483	return -1;
		1484
		1485	errno = ELOOP;
		1486	if (!--symlinks)
		1487	return -1;
		1488
		1489	if (*tmp1 == '/')
		1490	res = tmpbuf->ptr; /* symlink resolves to an absolute path */
		1491
		1492	/* we need to be careful, as rel might point into tmp2 already */
		1493	memmove (tmp2 + linklen + 1, rel, rellen + 1);
		1494	tmp2 [linklen] = '/';
		1495	memcpy (tmp2, tmp1, linklen);
		1496
		1497	rel = tmp2;
		1498	}
		1499	}
		1500
		1501	/* special case for the lone root path */
		1502	if (res == tmpbuf->ptr)
		1503	*res++ = '/';
		1504
		1505	return res - (char *)tmpbuf->ptr;
		1506	}
		1507
		1508	static signed char
		1509	eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
		1510	{
		1511	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
		1512	: a->inode < b->inode ? -1
		1513	: a->inode > b->inode ? 1
		1514	: 0;
		1515	}
		1516
		1517	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
		1518
		1519	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1520	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
		1521
		1522	static void
		1523	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
		1524	{
		1525	unsigned char bits [9 + sizeof (eio_ino_t) * 8];
		1526	unsigned char *bit = bits;
		1527
		1528	assert (CHAR_BIT == 8);
		1529	assert (sizeof (eio_dirent) * 8 < 256);
		1530	assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
		1531	assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
		1532
		1533	if (size <= EIO_SORT_FAST)
		1534	return;
		1535
		1536	/* first prepare an array of bits to test in our radix sort */
		1537	/* try to take endianness into account, as well as differences in eio_ino_t sizes */
		1538	/* inode_bits must contain all inodes ORed together */
		1539	/* which is used to skip bits that are 0 everywhere, which is very common */
		1540	{
		1541	eio_ino_t endianness;
		1542	int i, j;
		1543
		1544	/* we store the byte offset of byte n into byte n of "endianness" */
		1545	for (i = 0; i < sizeof (eio_ino_t); ++i)
		1546	((unsigned char *)&endianness)[i] = i;
		1547
		1548	*bit++ = 0;
		1549
		1550	for (i = 0; i < sizeof (eio_ino_t); ++i)
		1551	{
		1552	/* shifting off the byte offsets out of "endianness" */
		1553	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1554	endianness >>= 8;
		1555
		1556	for (j = 0; j < 8; ++j)
		1557	if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
		1558	*bit++ = offs + j;
		1559	}
		1560
		1561	for (j = 0; j < 8; ++j)
		1562	if (score_bits & (1 << j))
		1563	*bit++ = offsetof (eio_dirent, score) * 8 + j;
		1564	}
		1565
		1566	/* now actually do the sorting (a variant of MSD radix sort) */
		1567	{
		1568	eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
		1569	eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
		1570	unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
		1571	int stk_idx = 0;
		1572
		1573	base_stk [stk_idx] = dents;
		1574	end_stk [stk_idx] = dents + size;
		1575	bit_stk [stk_idx] = bit - 1;
		1576
		1577	do
		1578	{
		1579	base = base_stk [stk_idx];
		1580	end = end_stk [stk_idx];
		1581	bit = bit_stk [stk_idx];
		1582
		1583	for (;;)
		1584	{
		1585	unsigned char O = *bit >> 3;
		1586	unsigned char M = 1 << (*bit & 7);
		1587
		1588	eio_dirent *a = base;
		1589	eio_dirent *b = end;
		1590
		1591	if (b - a < EIO_SORT_CUTOFF)
746	res++;	1592	break;
747		1593
748	while (memofs + len > memlen)	1594	/* now bit-partition the array on the bit */
		1595	/* this ugly asymmetric loop seems to perform much better than typical */
		1596	/* partition algos found in the literature */
		1597	do
		1598	if (!(((unsigned char *)a)[O] & M))
		1599	++a;
		1600	else if (!(((unsigned char *)--b)[O] & M))
749	{	1601	{
750	memlen *= 2;	1602	eio_dirent tmp = *a; *a = *b; *b = tmp;
751	X_LOCK (wrklock);
752	req->ptr2 = names = realloc (names, memlen);
753	X_UNLOCK (wrklock);
754
755	if (!names)
756	break;	1603	++a;
757	}	1604	}
		1605	while (b > a);
758		1606
759	memcpy (names + memofs, name, len);	1607	/* next bit, or stop, if no bits left in this path */
760	memofs += len;	1608	if (!*--bit)
		1609	break;
		1610
		1611	base_stk [stk_idx] = a;
		1612	end_stk [stk_idx] = end;
		1613	bit_stk [stk_idx] = bit;
		1614	++stk_idx;
		1615
		1616	end = a;
761	}	1617	}
762	}	1618	}
763		1619	while (stk_idx--);
764	if (errno)
765	res = -1;
766		1620	}
		1621	}
		1622
		1623	static void
		1624	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1625	{
		1626	/* first move the smallest element to the front, to act as a sentinel */
		1627	{
		1628	int i;
		1629	eio_dirent *min = dents;
		1630
		1631	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1632	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1633	if (EIO_DENT_CMP (dents [i], <, *min))
		1634	min = &dents [i];
		1635
		1636	/* swap elements 0 and j (minimum) */
		1637	{
		1638	eio_dirent tmp = *dents; *dents = *min; *min = tmp;
		1639	}
		1640	}
		1641
		1642	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1643	{
		1644	eio_dirent *i, *j;
		1645
		1646	for (i = dents + 1; i < dents + size; ++i)
		1647	{
		1648	eio_dirent value = *i;
		1649
		1650	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
		1651	j [1] = j [0];
		1652
		1653	j [1] = value;
		1654	}
		1655	}
		1656	}
		1657
		1658	static void
		1659	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
		1660	{
		1661	if (size <= 1)
		1662	return; /* our insertion sort relies on size > 0 */
		1663
		1664	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1665	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1666	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1667
		1668	/* use an insertion sort at the end, or for small arrays, */
		1669	/* as insertion sort is more efficient for small partitions */
		1670	eio_dent_insertion_sort (dents, size);
		1671	}
		1672
		1673	/* read a full directory */
		1674	static void
		1675	eio__scandir (eio_req *req, etp_worker *self)
		1676	{
		1677	char *name, *names;
		1678	int namesalloc = 4096 - sizeof (void *) * 4;
		1679	int namesoffs = 0;
		1680	int flags = req->int1;
		1681	eio_dirent *dents = 0;
		1682	int dentalloc = 128;
		1683	int dentoffs = 0;
		1684	eio_ino_t inode_bits = 0;
		1685	#ifdef _WIN32
		1686	HANDLE dirp;
		1687	WIN32_FIND_DATA entp;
		1688	#else
		1689	DIR *dirp;
		1690	EIO_STRUCT_DIRENT *entp;
		1691	#endif
		1692
767	req->result = res;	1693	req->result = -1;
		1694
		1695	if (!(flags & EIO_READDIR_DENTS))
		1696	flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
		1697
		1698	#ifdef _WIN32
		1699	{
		1700	int len = strlen ((const char *)req->ptr1);
		1701	char *path = malloc (MAX_PATH);
		1702	const char *fmt;
		1703	const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
		1704
		1705	if (!len)
		1706	fmt = "./*";
		1707	else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
		1708	fmt = "%s*";
		1709	else
		1710	fmt = "%s/*";
		1711
		1712	_snprintf (path, MAX_PATH, fmt, reqpath);
		1713	dirp = FindFirstFile (path, &entp);
		1714	free (path);
		1715
		1716	if (dirp == INVALID_HANDLE_VALUE)
		1717	{
		1718	/* should steal _dosmaperr */
		1719	switch (GetLastError ())
		1720	{
		1721	case ERROR_FILE_NOT_FOUND:
		1722	req->result = 0;
		1723	break;
		1724
		1725	case ERROR_INVALID_NAME:
		1726	case ERROR_PATH_NOT_FOUND:
		1727	case ERROR_NO_MORE_FILES:
		1728	errno = ENOENT;
		1729	break;
		1730
		1731	case ERROR_NOT_ENOUGH_MEMORY:
		1732	errno = ENOMEM;
		1733	break;
		1734
		1735	default:
		1736	errno = EINVAL;
		1737	break;
		1738	}
		1739
		1740	return;
		1741	}
		1742	}
		1743	#else
		1744	#if HAVE_AT
		1745	if (req->wd)
		1746	{
		1747	int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
		1748
		1749	if (fd < 0)
		1750	return;
		1751
		1752	dirp = fdopendir (fd);
		1753
		1754	if (!dirp)
		1755	close (fd);
		1756	}
		1757	else
		1758	dirp = opendir (req->ptr1);
		1759	#else
		1760	dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
		1761	#endif
		1762
		1763	if (!dirp)
		1764	return;
		1765	#endif
		1766
		1767	if (req->flags & EIO_FLAG_PTR1_FREE)
		1768	free (req->ptr1);
		1769
		1770	req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
		1771	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
		1772	req->ptr2 = names = malloc (namesalloc);
		1773
		1774	if (!names || (flags && !dents))
		1775	return;
		1776
		1777	for (;;)
		1778	{
		1779	int done;
		1780
		1781	#ifdef _WIN32
		1782	done = !dirp;
		1783	#else
		1784	errno = 0;
		1785	entp = readdir (dirp);
		1786	done = !entp;
		1787	#endif
		1788
		1789	if (done)
		1790	{
		1791	#ifndef _WIN32
		1792	int old_errno = errno;
		1793	closedir (dirp);
		1794	errno = old_errno;
		1795
		1796	if (errno)
		1797	break;
		1798	#endif
		1799
		1800	/* sort etc. */
		1801	req->int1 = flags;
		1802	req->result = dentoffs;
		1803
		1804	if (flags & EIO_READDIR_STAT_ORDER)
		1805	eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
		1806	else if (flags & EIO_READDIR_DIRS_FIRST)
		1807	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1808	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1809	else
		1810	{
		1811	/* in this case, all is known, and we just put dirs first and sort them */
		1812	eio_dirent *oth = dents + dentoffs;
		1813	eio_dirent *dir = dents;
		1814
		1815	/* now partition dirs to the front, and non-dirs to the back */
		1816	/* by walking from both sides and swapping if necessary */
		1817	while (oth > dir)
		1818	{
		1819	if (dir->type == EIO_DT_DIR)
		1820	++dir;
		1821	else if ((--oth)->type == EIO_DT_DIR)
		1822	{
		1823	eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
		1824
		1825	++dir;
		1826	}
		1827	}
		1828
		1829	/* now sort the dirs only (dirs all have the same score) */
		1830	eio_dent_sort (dents, dir - dents, 0, inode_bits);
		1831	}
		1832
		1833	break;
		1834	}
		1835
		1836	/* now add the entry to our list(s) */
		1837	name = D_NAME (entp);
		1838
		1839	/* skip . and .. entries */
		1840	if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
		1841	{
		1842	int len = D_NAMLEN (entp) + 1;
		1843
		1844	while (ecb_expect_false (namesoffs + len > namesalloc))
		1845	{
		1846	namesalloc *= 2;
		1847	req->ptr2 = names = realloc (names, namesalloc);
		1848
		1849	if (!names)
		1850	break;
		1851	}
		1852
		1853	memcpy (names + namesoffs, name, len);
		1854
		1855	if (dents)
		1856	{
		1857	struct eio_dirent *ent;
		1858
		1859	if (ecb_expect_false (dentoffs == dentalloc))
		1860	{
		1861	dentalloc *= 2;
		1862	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1863
		1864	if (!dents)
		1865	break;
		1866	}
		1867
		1868	ent = dents + dentoffs;
		1869
		1870	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
		1871	ent->namelen = len - 1;
		1872	ent->inode = D_INO (entp);
		1873
		1874	inode_bits |= ent->inode;
		1875
		1876	switch (D_TYPE (entp))
		1877	{
		1878	default:
		1879	ent->type = EIO_DT_UNKNOWN;
		1880	flags |= EIO_READDIR_FOUND_UNKNOWN;
		1881	break;
		1882
		1883	#ifdef DT_FIFO
		1884	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1885	#endif
		1886	#ifdef DT_CHR
		1887	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1888	#endif
		1889	#ifdef DT_MPC
		1890	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1891	#endif
		1892	#ifdef DT_DIR
		1893	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1894	#endif
		1895	#ifdef DT_NAM
		1896	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1897	#endif
		1898	#ifdef DT_BLK
		1899	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1900	#endif
		1901	#ifdef DT_MPB
		1902	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1903	#endif
		1904	#ifdef DT_REG
		1905	case DT_REG: ent->type = EIO_DT_REG; break;
		1906	#endif
		1907	#ifdef DT_NWK
		1908	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1909	#endif
		1910	#ifdef DT_CMP
		1911	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1912	#endif
		1913	#ifdef DT_LNK
		1914	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1915	#endif
		1916	#ifdef DT_SOCK
		1917	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1918	#endif
		1919	#ifdef DT_DOOR
		1920	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1921	#endif
		1922	#ifdef DT_WHT
		1923	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1924	#endif
		1925	}
		1926
		1927	ent->score = 7;
		1928
		1929	if (flags & EIO_READDIR_DIRS_FIRST)
		1930	{
		1931	if (ent->type == EIO_DT_UNKNOWN)
		1932	{
		1933	if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
		1934	ent->score = 1;
		1935	else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
		1936	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
		1937	}
		1938	else if (ent->type == EIO_DT_DIR)
		1939	ent->score = 0;
		1940	}
		1941	}
		1942
		1943	namesoffs += len;
		1944	++dentoffs;
		1945	}
		1946
		1947	if (EIO_CANCELLED (req))
		1948	{
		1949	errno = ECANCELED;
		1950	break;
		1951	}
		1952
		1953	#ifdef _WIN32
		1954	if (!FindNextFile (dirp, &entp))
		1955	{
		1956	FindClose (dirp);
		1957	dirp = 0;
		1958	}
		1959	#endif
		1960	}
768	}	1961	}
		1962
		1963	/*****************************************************************************/
		1964	/* working directory stuff */
		1965	/* various deficiencies in the posix 2008 api force us to */
		1966	/* keep the absolute path in string form at all times */
		1967	/* fuck yeah. */
		1968
		1969	#if !HAVE_AT
		1970
		1971	/* a bit like realpath, but usually faster because it doesn'T have to return */
		1972	/* an absolute or canonical path */
		1973	static const char *
		1974	wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
		1975	{
		1976	if (!wd || *path == '/')
		1977	return path;
		1978
		1979	if (path [0] == '.' && !path [1])
		1980	return wd->str;
		1981
		1982	{
		1983	int l1 = wd->len;
		1984	int l2 = strlen (path);
		1985
		1986	char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
		1987
		1988	memcpy (res, wd->str, l1);
		1989	res [l1] = '/';
		1990	memcpy (res + l1 + 1, path, l2 + 1);
		1991
		1992	return res;
		1993	}
		1994	}
		1995
		1996	#endif
		1997
		1998	static eio_wd
		1999	eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
		2000	{
		2001	int fd;
		2002	eio_wd res;
		2003	int len = eio__realpath (tmpbuf, wd, path);
		2004
		2005	if (len < 0)
		2006	return EIO_INVALID_WD;
		2007
		2008	#if HAVE_AT
		2009	fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
		2010
		2011	if (fd < 0)
		2012	return EIO_INVALID_WD;
		2013	#endif
		2014
		2015	res = malloc (sizeof (*res) + len); /* one extra 0-byte */
		2016
		2017	#if HAVE_AT
		2018	res->fd = fd;
		2019	#endif
		2020
		2021	res->len = len;
		2022	memcpy (res->str, tmpbuf->ptr, len);
		2023	res->str [len] = 0;
		2024
		2025	return res;
		2026	}
		2027
		2028	eio_wd
		2029	eio_wd_open_sync (eio_wd wd, const char *path)
		2030	{
		2031	struct tmpbuf tmpbuf = { 0 };
		2032	wd = eio__wd_open_sync (&tmpbuf, wd, path);
		2033	free (tmpbuf.ptr);
		2034
		2035	return wd;
		2036	}
		2037
		2038	void
		2039	eio_wd_close_sync (eio_wd wd)
		2040	{
		2041	if (wd != EIO_INVALID_WD && wd != EIO_CWD)
		2042	{
		2043	#if HAVE_AT
		2044	close (wd->fd);
		2045	#endif
		2046	free (wd);
		2047	}
		2048	}
		2049
		2050	#if HAVE_AT
		2051
		2052	/* they forgot these */
		2053
		2054	static int
		2055	eio__truncateat (int dirfd, const char *path, off_t length)
		2056	{
		2057	int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
		2058	int res;
		2059
		2060	if (fd < 0)
		2061	return fd;
		2062
		2063	res = ftruncate (fd, length);
		2064	close (fd);
		2065	return res;
		2066	}
		2067
		2068	static int
		2069	eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
		2070	{
		2071	int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
		2072	int res;
		2073
		2074	if (fd < 0)
		2075	return fd;
		2076
		2077	res = fstatvfs (fd, buf);
		2078	close (fd);
		2079	return res;
		2080
		2081	}
		2082
		2083	#endif
769		2084
770	/*****************************************************************************/	2085	/*****************************************************************************/
771		2086
772	#define ALLOC(len) \	2087	#define ALLOC(len) \
773	if (!req->ptr2) \	2088	if (!req->ptr2) \
…		…
782	req->result = -1; \	2097	req->result = -1; \
783	break; \	2098	break; \
784	} \	2099	} \
785	}	2100	}
786		2101
		2102	static void ecb_noinline ecb_cold
		2103	etp_proc_init (void)
		2104	{
		2105	#if HAVE_PRCTL_SET_NAME
		2106	/* provide a more sensible "thread name" */
		2107	char name[16 + 1];
		2108	const int namelen = sizeof (name) - 1;
		2109	int len;
		2110
		2111	prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0);
		2112	name [namelen] = 0;
		2113	len = strlen (name);
		2114	strcpy (name + (len <= namelen - 4 ? len : namelen - 4), "/eio");
		2115	prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0);
		2116	#endif
		2117	}
		2118
787	X_THREAD_PROC (eio_proc)	2119	X_THREAD_PROC (etp_proc)
788	{	2120	{
789	eio_req *req;	2121	ETP_REQ *req;
790	struct timespec ts;	2122	struct timespec ts;
791	worker *self = (worker *)thr_arg;	2123	etp_worker *self = (etp_worker *)thr_arg;
792		2124
		2125	etp_proc_init ();
		2126
793	/* try to distribute timeouts somewhat randomly */	2127	/* try to distribute timeouts somewhat evenly */
794	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);	2128	ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
795		2129
796	for (;;)	2130	for (;;)
797	{	2131	{
798	ts.tv_sec = time (0) + IDLE_TIMEOUT;	2132	ts.tv_sec = 0;
799		2133
800	X_LOCK (reqlock);	2134	X_LOCK (reqlock);
801		2135
802	for (;;)	2136	for (;;)
803	{	2137	{
804	self->req = req = reqq_shift (&req_queue);	2138	req = reqq_shift (&req_queue);
805		2139
806	if (req)	2140	if (req)
807	break;	2141	break;
808		2142
		2143	if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
		2144	{
		2145	X_UNLOCK (reqlock);
		2146	X_LOCK (wrklock);
		2147	--started;
		2148	X_UNLOCK (wrklock);
		2149	goto quit;
		2150	}
		2151
809	++idle;	2152	++idle;
810		2153
811	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)	2154	if (idle <= max_idle)
		2155	/* we are allowed to idle, so do so without any timeout */
		2156	X_COND_WAIT (reqwait, reqlock);
		2157	else
812	{	2158	{
813	if (idle > max_idle)	2159	/* initialise timeout once */
814	{	2160	if (!ts.tv_sec)
815	--idle;	2161	ts.tv_sec = time (0) + idle_timeout;
816	X_UNLOCK (reqlock);
817	X_LOCK (wrklock);
818	--started;
819	X_UNLOCK (wrklock);
820	goto quit;
821	}
822		2162
823	/* we are allowed to idle, so do so without any timeout */
824	X_COND_WAIT (reqwait, reqlock);	2163	if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
825	ts.tv_sec = time (0) + IDLE_TIMEOUT;	2164	ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
826	}	2165	}
827		2166
828	--idle;	2167	--idle;
829	}	2168	}
830		2169
831	--nready;	2170	--nready;
832		2171
833	X_UNLOCK (reqlock);	2172	X_UNLOCK (reqlock);
834		2173
835	errno = 0; /* strictly unnecessary */
836
837	if (!EIO_CANCELLED (req))
838	switch (req->type)	2174	if (req->type < 0)
839	{
840	case EIO_READ: ALLOC (req->size);
841	req->result = req->offs >= 0
842	? pread (req->int1, req->ptr2, req->size, req->offs)
843	: read (req->int1, req->ptr2, req->size); break;
844	case EIO_WRITE: req->result = req->offs >= 0
845	? pwrite (req->int1, req->ptr2, req->size, req->offs)
846	: write (req->int1, req->ptr2, req->size); break;
847
848	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
849	case EIO_SENDFILE: req->result = sendfile_ (req->int1, req->int2, req->offs, req->size, self); break;
850
851	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
852	req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
853	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
854	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
855	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
856	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
857
858	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
859	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
860	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
861	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
862	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
863	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
864
865	case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
866	case EIO_CLOSE: req->result = close (req->int1); break;
867	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
868	case EIO_UNLINK: req->result = unlink (req->ptr1); break;
869	case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
870	case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
871	case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
872	case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
873	case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
874	case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
875
876	case EIO_READLINK: ALLOC (NAME_MAX);
877	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
878
879	case EIO_SYNC: req->result = 0; sync (); break;
880	case EIO_FSYNC: req->result = fsync (req->int1); break;
881	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
882
883	case EIO_READDIR: scandir_ (req, self); break;
884
885	case EIO_BUSY:
886	#ifdef _WIN32
887	Sleep (req->nv1 * 1000.);
888	#else
889	{
890	struct timeval tv;
891
892	tv.tv_sec = req->nv1;
893	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
894
895	req->result = select (0, 0, 0, 0, &tv);
896	}
897	#endif
898	break;
899
900	case EIO_UTIME:
901	case EIO_FUTIME:
902	{
903	struct timeval tv[2];
904	struct timeval *times;
905
906	if (req->nv1 != -1. || req->nv2 != -1.)
907	{
908	tv[0].tv_sec = req->nv1;
909	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
910	tv[1].tv_sec = req->nv2;
911	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
912
913	times = tv;
914	}
915	else
916	times = 0;
917
918
919	req->result = req->type == EIO_FUTIME
920	? futimes (req->int1, times)
921	: utimes (req->ptr1, times);
922	}
923
924	case EIO_GROUP:
925	case EIO_NOP:
926	break;
927
928	case EIO_QUIT:
929	goto quit;	2175	goto quit;
930		2176
931	default:	2177	ETP_EXECUTE (self, req);
932	req->result = -1;
933	break;
934	}
935
936	req->errorno = errno;
937		2178
938	X_LOCK (reslock);	2179	X_LOCK (reslock);
939		2180
940	++npending;	2181	++npending;
941		2182
942	if (!reqq_push (&res_queue, req) && want_poll_cb)	2183	if (!reqq_push (&res_queue, req) && want_poll_cb)
943	want_poll_cb ();	2184	want_poll_cb ();
944		2185
945	self->req = 0;
946	worker_clear (self);	2186	etp_worker_clear (self);
947		2187
948	X_UNLOCK (reslock);	2188	X_UNLOCK (reslock);
949	}	2189	}
950		2190
951	quit:	2191	quit:
		2192	free (req);
		2193
952	X_LOCK (wrklock);	2194	X_LOCK (wrklock);
953	worker_free (self);	2195	etp_worker_free (self);
954	X_UNLOCK (wrklock);	2196	X_UNLOCK (wrklock);
955
956	return 0;
957	}	2197	}
958		2198
959	/*****************************************************************************/	2199	/*****************************************************************************/
960		2200
961	static void eio_atfork_prepare (void)	2201	int ecb_cold
962	{
963	X_LOCK (wrklock);
964	X_LOCK (reqlock);
965	X_LOCK (reslock);
966	#if !HAVE_PREADWRITE
967	X_LOCK (preadwritelock);
968	#endif
969	#if !HAVE_READDIR_R
970	X_LOCK (readdirlock);
971	#endif
972	}
973
974	static void eio_atfork_parent (void)
975	{
976	#if !HAVE_READDIR_R
977	X_UNLOCK (readdirlock);
978	#endif
979	#if !HAVE_PREADWRITE
980	X_UNLOCK (preadwritelock);
981	#endif
982	X_UNLOCK (reslock);
983	X_UNLOCK (reqlock);
984	X_UNLOCK (wrklock);
985	}
986
987	static void eio_atfork_child (void)
988	{
989	eio_req *prv;
990
991	while (prv = reqq_shift (&req_queue))
992	eio_destroy (prv);
993
994	while (prv = reqq_shift (&res_queue))
995	eio_destroy (prv);
996
997	while (wrk_first.next != &wrk_first)
998	{
999	worker *wrk = wrk_first.next;
1000
1001	if (wrk->req)
1002	eio_destroy (wrk->req);
1003
1004	worker_clear (wrk);
1005	worker_free (wrk);
1006	}
1007
1008	started = 0;
1009	idle = 0;
1010	nreqs = 0;
1011	nready = 0;
1012	npending = 0;
1013
1014	eio_atfork_parent ();
1015	}
1016
1017	int eio_init (void (*want_poll)(void), void (*done_poll)(void))	2202	eio_init (void (*want_poll)(void), void (*done_poll)(void))
1018	{	2203	{
1019	want_poll_cb = want_poll;	2204	return etp_init (want_poll, done_poll);
1020	done_poll_cb = done_poll;
1021
1022	#ifdef _WIN32
1023	X_MUTEX_CHECK (wrklock);
1024	X_MUTEX_CHECK (reslock);
1025	X_MUTEX_CHECK (reqlock);
1026	X_MUTEX_CHECK (reqwait);
1027	X_MUTEX_CHECK (preadwritelock);
1028	X_MUTEX_CHECK (readdirlock);
1029
1030	X_COND_CHECK (reqwait);
1031	#endif
1032
1033	X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child);
1034	}	2205	}
1035		2206
		2207	ecb_inline void
1036	static void eio_api_destroy (eio_req *req)	2208	eio_api_destroy (eio_req *req)
1037	{	2209	{
1038	free (req);	2210	free (req);
1039	}	2211	}
1040		2212
1041	#define REQ(rtype) \	2213	#define REQ(rtype) \
1042	eio_req *req; \	2214	eio_req *req; \
1043	\	2215	\
1044	req = (eio_req *)calloc (1, sizeof *req); \	2216	req = (eio_req *)calloc (1, sizeof *req); \
1045	if (!req) \	2217	if (!req) \
1046	return 0; \	2218	return 0; \
1047	\	2219	\
1048	req->type = EIO_ ## rtype; \	2220	req->type = rtype; \
1049	req->pri = EIO_DEFAULT_PRI + EIO_PRI_BIAS; \	2221	req->pri = pri; \
1050	req->finish = cb; \	2222	req->finish = cb; \
		2223	req->data = data; \
1051	req->destroy = eio_api_destroy;	2224	req->destroy = eio_api_destroy;
1052		2225
1053	#define SEND eio_submit (req); return req	2226	#define SEND eio_submit (req); return req
1054		2227
1055	#define PATH (void)0	2228	#define PATH \
		2229	req->flags |= EIO_FLAG_PTR1_FREE; \
		2230	req->ptr1 = strdup (path); \
		2231	if (!req->ptr1) \
		2232	{ \
		2233	eio_api_destroy (req); \
		2234	return 0; \
		2235	}
1056		2236
		2237	static void
		2238	eio_execute (etp_worker *self, eio_req *req)
		2239	{
		2240	#if HAVE_AT
		2241	int dirfd;
		2242	#else
		2243	const char *path;
		2244	#endif
		2245
		2246	if (ecb_expect_false (EIO_CANCELLED (req)))
		2247	{
		2248	req->result = -1;
		2249	req->errorno = ECANCELED;
		2250	return;
		2251	}
		2252
		2253	if (ecb_expect_false (req->wd == EIO_INVALID_WD))
		2254	{
		2255	req->result = -1;
		2256	req->errorno = ENOENT;
		2257	return;
		2258	}
		2259
		2260	if (req->type >= EIO_OPEN)
		2261	{
		2262	#if HAVE_AT
		2263	dirfd = WD2FD (req->wd);
		2264	#else
		2265	path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
		2266	#endif
		2267	}
		2268
		2269	switch (req->type)
		2270	{
		2271	case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
		2272	req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
		2273	break;
		2274	case EIO_WD_CLOSE: req->result = 0;
		2275	eio_wd_close_sync (req->wd); break;
		2276
		2277	case EIO_SEEK: eio__lseek (req); break;
		2278	case EIO_READ: ALLOC (req->size);
		2279	req->result = req->offs >= 0
		2280	? pread (req->int1, req->ptr2, req->size, req->offs)
		2281	: read (req->int1, req->ptr2, req->size); break;
		2282	case EIO_WRITE: req->result = req->offs >= 0
		2283	? pwrite (req->int1, req->ptr2, req->size, req->offs)
		2284	: write (req->int1, req->ptr2, req->size); break;
		2285
		2286	case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
		2287	case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
		2288
		2289	#if HAVE_AT
		2290
		2291	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		2292	req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
		2293	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		2294	req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
		2295	case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
		2296	case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
		2297	case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
		2298	case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
		2299
		2300	case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
		2301	case EIO_RMDIR: req->result = unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
		2302	case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
		2303	case EIO_RENAME: req->result = renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
		2304	case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
		2305	case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
		2306	case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
		2307	case EIO_READLINK: ALLOC (PATH_MAX);
		2308	req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
		2309	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2310	req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		2311	case EIO_UTIME:
		2312	case EIO_FUTIME:
		2313	{
		2314	struct timespec ts[2];
		2315	struct timespec *times;
		2316
		2317	if (req->nv1 != -1. || req->nv2 != -1.)
		2318	{
		2319	ts[0].tv_sec = req->nv1;
		2320	ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
		2321	ts[1].tv_sec = req->nv2;
		2322	ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
		2323
		2324	times = ts;
		2325	}
		2326	else
		2327	times = 0;
		2328
		2329	req->result = req->type == EIO_FUTIME
		2330	? futimens (req->int1, times)
		2331	: utimensat (dirfd, req->ptr1, times, 0);
		2332	}
		2333	break;
		2334
		2335	#else
		2336
		2337	case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		2338	req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
		2339	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		2340	req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
		2341	case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
		2342	case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
		2343	case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
		2344	case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
		2345
		2346	case EIO_UNLINK: req->result = unlink (path ); break;
		2347	case EIO_RMDIR: req->result = rmdir (path ); break;
		2348	case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
		2349	case EIO_RENAME: req->result = rename (path , req->ptr2); break;
		2350	case EIO_LINK: req->result = link (path , req->ptr2); break;
		2351	case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
		2352	case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
		2353	case EIO_READLINK: ALLOC (PATH_MAX);
		2354	req->result = readlink (path, req->ptr2, PATH_MAX); break;
		2355	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2356	req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
		2357
		2358	case EIO_UTIME:
		2359	case EIO_FUTIME:
		2360	{
		2361	struct timeval tv[2];
		2362	struct timeval *times;
		2363
		2364	if (req->nv1 != -1. || req->nv2 != -1.)
		2365	{
		2366	tv[0].tv_sec = req->nv1;
		2367	tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
		2368	tv[1].tv_sec = req->nv2;
		2369	tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
		2370
		2371	times = tv;
		2372	}
		2373	else
		2374	times = 0;
		2375
		2376	req->result = req->type == EIO_FUTIME
		2377	? futimes (req->int1, times)
		2378	: utimes (req->ptr1, times);
		2379	}
		2380	break;
		2381
		2382	#endif
		2383
		2384	case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
		2385	{
		2386	ALLOC (req->result);
		2387	memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
		2388	}
		2389	break;
		2390
		2391	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
		2392	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
		2393
		2394	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		2395	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		2396
		2397	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
		2398	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
		2399	case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
		2400
		2401	case EIO_CLOSE: req->result = close (req->int1); break;
		2402	case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
		2403	case EIO_SYNC: req->result = 0; sync (); break;
		2404	case EIO_FSYNC: req->result = fsync (req->int1); break;
		2405	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
		2406	case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
		2407	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
		2408	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
		2409	case EIO_MTOUCH: req->result = eio__mtouch (req); break;
		2410	case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
		2411	case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
		2412	case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
		2413
		2414	case EIO_READDIR: eio__scandir (req, self); break;
		2415
		2416	case EIO_BUSY:
		2417	#ifdef _WIN32
		2418	Sleep (req->nv1 * 1e3);
		2419	#else
		2420	{
		2421	struct timeval tv;
		2422
		2423	tv.tv_sec = req->nv1;
		2424	tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
		2425
		2426	req->result = select (0, 0, 0, 0, &tv);
		2427	}
		2428	#endif
		2429	break;
		2430
		2431	case EIO_GROUP:
		2432	abort (); /* handled in eio_request */
		2433
		2434	case EIO_NOP:
		2435	req->result = 0;
		2436	break;
		2437
		2438	case EIO_CUSTOM:
		2439	req->feed (req);
		2440	break;
		2441
		2442	default:
		2443	req->result = EIO_ENOSYS ();
		2444	break;
		2445	}
		2446
		2447	req->errorno = errno;
		2448	}
		2449
		2450	#ifndef EIO_NO_WRAPPERS
		2451
		2452	eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
		2453	{
		2454	REQ (EIO_WD_OPEN); PATH; SEND;
		2455	}
		2456
		2457	eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
		2458	{
		2459	REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
		2460	}
		2461
		2462	eio_req *eio_nop (int pri, eio_cb cb, void *data)
		2463	{
		2464	REQ (EIO_NOP); SEND;
		2465	}
		2466
		2467	eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
		2468	{
		2469	REQ (EIO_BUSY); req->nv1 = delay; SEND;
		2470	}
		2471
1057	eio_req *eio_fsync (int fd, eio_cb cb)	2472	eio_req *eio_sync (int pri, eio_cb cb, void *data)
1058	{	2473	{
		2474	REQ (EIO_SYNC); SEND;
		2475	}
		2476
		2477	eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
		2478	{
1059	REQ (FSYNC); req->int1 = fd; SEND;	2479	REQ (EIO_FSYNC); req->int1 = fd; SEND;
1060	}	2480	}
1061		2481
		2482	eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		2483	{
		2484	REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		2485	}
		2486
1062	eio_req *eio_fdatasync (int fd, eio_cb cb)	2487	eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1063	{	2488	{
1064	REQ (FDATASYNC); req->int1 = fd; SEND;	2489	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1065	}	2490	}
1066		2491
		2492	eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
		2493	{
		2494	REQ (EIO_SYNCFS); req->int1 = fd; SEND;
		2495	}
		2496
		2497	eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
		2498	{
		2499	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		2500	}
		2501
		2502	eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
		2503	{
		2504	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
		2505	}
		2506
		2507	eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
		2508	{
		2509	REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
		2510	}
		2511
		2512	eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
		2513	{
		2514	REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
		2515	}
		2516
		2517	eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
		2518	{
		2519	REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
		2520	}
		2521
		2522	eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
		2523	{
		2524	REQ (EIO_CLOSE); req->int1 = fd; SEND;
		2525	}
		2526
1067	eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb)	2527	eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1068	{	2528	{
1069	REQ (READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;	2529	REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1070	}	2530	}
1071		2531
1072	eio_req *eio_read (int fd, off_t offset, size_t length, void *data, eio_cb cb)	2532	eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
1073	{	2533	{
		2534	REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
		2535	}
		2536
		2537	eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
		2538	{
1074	REQ (READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = data; SEND;	2539	REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1075	}	2540	}
1076		2541
1077	eio_req *eio_write (int fd, off_t offset, size_t length, void *data, eio_cb cb)	2542	eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1078	{	2543	{
1079	REQ (WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = data; SEND;	2544	REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1080	}	2545	}
1081		2546
1082	eio_req *eio_fstat (int fd, eio_cb cb)	2547	eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1083	{	2548	{
1084	REQ (FSTAT); req->int1 = fd; SEND;	2549	REQ (EIO_FSTAT); req->int1 = fd; SEND;
1085	}	2550	}
1086		2551
		2552	eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
		2553	{
		2554	REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
		2555	}
		2556
1087	eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb)	2557	eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1088	{	2558	{
1089	REQ (FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;	2559	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1090	}	2560	}
1091		2561
1092	eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb)	2562	eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
1093	{	2563	{
1094	REQ (FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;	2564	REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
1095	}	2565	}
1096		2566
1097	eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb)	2567	eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1098	{	2568	{
1099	REQ (FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;	2569	REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1100	}	2570	}
1101		2571
1102	eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb)	2572	eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1103	{	2573	{
1104	REQ (FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2574	REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1105	}	2575	}
1106		2576
1107	eio_req *eio_dup2 (int fd, int fd2, eio_cb cb)	2577	eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1108	{	2578	{
1109	REQ (DUP2); req->int1 = fd; req->int2 = fd2; SEND;	2579	REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
1110	}	2580	}
1111		2581
1112	eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb)	2582	eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
1113	{	2583	{
1114	REQ (SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;	2584	REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
1115	}	2585	}
1116		2586
1117	eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb)	2587	eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
1118	{	2588	{
1119	REQ (OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;	2589	REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
1120	}	2590	}
1121		2591
1122	eio_req *eio_readlink (const char *path, eio_cb cb)
1123	{
1124	REQ (READLINK); PATH; SEND;
1125	}
1126
1127	eio_req *eio_stat (const char *path, eio_cb cb)
1128	{
1129	REQ (STAT); PATH; SEND;
1130	}
1131
1132	eio_req *eio_lstat (const char *path, eio_cb cb)
1133	{
1134	REQ (LSTAT); PATH; SEND;
1135	}
1136
1137	eio_req *eio_utime (const char *path, double atime, double mtime, eio_cb cb)	2592	eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
1138	{	2593	{
1139	REQ (UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;	2594	REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
1140	}	2595	}
1141		2596
1142	eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb)	2597	eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1143	{	2598	{
1144	REQ (TRUNCATE); PATH; req->offs = offset; SEND;	2599	REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1145	}	2600	}
1146		2601
1147	eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb)	2602	eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1148	{	2603	{
1149	REQ (CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;	2604	REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1150	}	2605	}
1151		2606
1152	eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb)	2607	eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1153	{	2608	{
1154	REQ (CHMOD); PATH; req->int2 = (long)mode; SEND;	2609	REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
1155	}	2610	}
1156		2611
1157	eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb)	2612	eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1158	{	2613	{
1159	REQ (MKDIR); PATH; req->int2 = (long)mode; SEND;	2614	REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
1160	}	2615	}
1161		2616
1162	eio_req *eio_unlink (const char *path, eio_cb cb)	2617	static eio_req *
		2618	eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
1163	{	2619	{
1164	REQ (UNLINK); PATH; SEND;
1165	}
1166
1167	eio_req *eio_rmdir (const char *path, eio_cb cb)
1168	{
1169	REQ (RMDIR); PATH; SEND;	2620	REQ (type); PATH; SEND;
1170	}	2621	}
1171		2622
		2623	eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
		2624	{
		2625	return eio__1path (EIO_READLINK, path, pri, cb, data);
		2626	}
		2627
		2628	eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
		2629	{
		2630	return eio__1path (EIO_REALPATH, path, pri, cb, data);
		2631	}
		2632
		2633	eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
		2634	{
		2635	return eio__1path (EIO_STAT, path, pri, cb, data);
		2636	}
		2637
		2638	eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
		2639	{
		2640	return eio__1path (EIO_LSTAT, path, pri, cb, data);
		2641	}
		2642
		2643	eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
		2644	{
		2645	return eio__1path (EIO_STATVFS, path, pri, cb, data);
		2646	}
		2647
		2648	eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
		2649	{
		2650	return eio__1path (EIO_UNLINK, path, pri, cb, data);
		2651	}
		2652
1172	eio_req *eio_readdir (const char *path, eio_cb cb)	2653	eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1173	{	2654	{
1174	REQ (READDIR); PATH; SEND;	2655	return eio__1path (EIO_RMDIR, path, pri, cb, data);
1175	}	2656	}
1176		2657
		2658	eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
		2659	{
		2660	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
		2661	}
		2662
1177	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb)	2663	eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1178	{	2664	{
1179	REQ (MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND;	2665	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1180	}	2666	}
1181		2667
1182	eio_req *eio_busy (double delay, eio_cb cb)	2668	static eio_req *
		2669	eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1183	{	2670	{
1184	REQ (BUSY); req->nv1 = delay; SEND;	2671	REQ (type); PATH;
1185	}
1186		2672
		2673	req->flags |= EIO_FLAG_PTR2_FREE;
		2674	req->ptr2 = strdup (new_path);
		2675	if (!req->ptr2)
		2676	{
		2677	eio_api_destroy (req);
		2678	return 0;
		2679	}
		2680
		2681	SEND;
		2682	}
		2683
		2684	eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
		2685	{
		2686	return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
		2687	}
		2688
		2689	eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
		2690	{
		2691	return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
		2692	}
		2693
		2694	eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
		2695	{
		2696	return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
		2697	}
		2698
		2699	eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
		2700	{
		2701	REQ (EIO_CUSTOM); req->feed = execute; SEND;
		2702	}
		2703
		2704	#endif
		2705
1187	eio_req *eio_nop (eio_cb cb)	2706	eio_req *eio_grp (eio_cb cb, void *data)
1188	{	2707	{
		2708	const int pri = EIO_PRI_MAX;
		2709
1189	REQ (NOP); SEND;	2710	REQ (EIO_GROUP); SEND;
1190	}	2711	}
1191		2712
1192	#undef REQ	2713	#undef REQ
1193	#undef PATH	2714	#undef PATH
1194	#undef SEND	2715	#undef SEND
1195		2716
1196	#if 0	2717	/*****************************************************************************/
		2718	/* grp functions */
1197		2719
1198	void	2720	void
1199	aio_open (SV8 *pathname, int flags, int mode, SV *callback=&PL_sv_undef)
1200	PROTOTYPE: $$$;$
1201	PPCODE:
1202	{
1203	dREQ;
1204
1205	req->type = EIO_OPEN;
1206	req->sv1 = newSVsv (pathname);
1207	req->ptr1 = SvPVbyte_nolen (req->sv1);
1208	req->int1 = flags;
1209	req->int2 = mode;
1210
1211	EIO_SEND;
1212	}
1213
1214	void
1215	aio_fsync (SV *fh, SV *callback=&PL_sv_undef)
1216	PROTOTYPE: $;$
1217	ALIAS:
1218	aio_fsync = EIO_FSYNC
1219	aio_fdatasync = EIO_FDATASYNC
1220	PPCODE:
1221	{
1222	dREQ;
1223
1224	req->type = ix;
1225	req->sv1 = newSVsv (fh);
1226	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1227
1228	EIO_SEND (req);
1229	}
1230
1231	void
1232	aio_close (SV *fh, SV *callback=&PL_sv_undef)
1233	PROTOTYPE: $;$
1234	PPCODE:
1235	{
1236	dREQ;
1237
1238	req->type = EIO_CLOSE;
1239	req->sv1 = newSVsv (fh);
1240	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1241
1242	EIO_SEND (req);
1243	}
1244
1245	void
1246	aio_read (SV *fh, SV *offset, SV *length, SV8 *data, IV dataoffset, SV *callback=&PL_sv_undef)
1247	ALIAS:
1248	aio_read = EIO_READ
1249	aio_write = EIO_WRITE
1250	PROTOTYPE: $$$$$;$
1251	PPCODE:
1252	{
1253	STRLEN svlen;
1254	char *svptr = SvPVbyte (data, svlen);
1255	UV len = SvUV (length);
1256
1257	SvUPGRADE (data, SVt_PV);
1258	SvPOK_on (data);
1259
1260	if (dataoffset < 0)
1261	dataoffset += svlen;
1262
1263	if (dataoffset < 0 || dataoffset > svlen)
1264	croak ("dataoffset outside of data scalar");
1265
1266	if (ix == EIO_WRITE)
1267	{
1268	/* write: check length and adjust. */
1269	if (!SvOK (length) || len + dataoffset > svlen)
1270	len = svlen - dataoffset;
1271	}
1272	else
1273	{
1274	/* read: grow scalar as necessary */
1275	svptr = SvGROW (data, len + dataoffset + 1);
1276	}
1277
1278	if (len < 0)
1279	croak ("length must not be negative");
1280
1281	{
1282	dREQ;
1283
1284	req->type = ix;
1285	req->sv1 = newSVsv (fh);
1286	req->int1 = PerlIO_fileno (ix == EIO_READ ? IoIFP (sv_2io (fh))
1287	: IoOFP (sv_2io (fh)));
1288	req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1289	req->size = len;
1290	req->sv2 = SvREFCNT_inc (data);
1291	req->ptr2 = (char *)svptr + dataoffset;
1292	req->stroffset = dataoffset;
1293
1294	if (!SvREADONLY (data))
1295	{
1296	SvREADONLY_on (data);
1297	req->flags |= FLAG_SV2_RO_OFF;
1298	}
1299
1300	EIO_SEND;
1301	}
1302	}
1303
1304	void
1305	aio_readlink (SV8 *path, SV *callback=&PL_sv_undef)
1306	PROTOTYPE: $$;$
1307	PPCODE:
1308	{
1309	SV *data;
1310	dREQ;
1311
1312	data = newSV (NAME_MAX);
1313	SvPOK_on (data);
1314
1315	req->type = EIO_READLINK;
1316	req->sv1 = newSVsv (path);
1317	req->ptr1 = SvPVbyte_nolen (req->sv1);
1318	req->sv2 = data;
1319	req->ptr2 = SvPVbyte_nolen (data);
1320
1321	EIO_SEND;
1322	}
1323
1324	void
1325	aio_sendfile (SV *out_fh, SV *in_fh, SV *in_offset, UV length, SV *callback=&PL_sv_undef)
1326	PROTOTYPE: $$$$;$
1327	PPCODE:
1328	{
1329	dREQ;
1330
1331	req->type = EIO_SENDFILE;
1332	req->sv1 = newSVsv (out_fh);
1333	req->int1 = PerlIO_fileno (IoIFP (sv_2io (out_fh)));
1334	req->sv2 = newSVsv (in_fh);
1335	req->int2 = PerlIO_fileno (IoIFP (sv_2io (in_fh)));
1336	req->offs = SvVAL64 (in_offset);
1337	req->size = length;
1338
1339	EIO_SEND;
1340	}
1341
1342	void
1343	aio_readahead (SV *fh, SV *offset, IV length, SV *callback=&PL_sv_undef)
1344	PROTOTYPE: $$$;$
1345	PPCODE:
1346	{
1347	dREQ;
1348
1349	req->type = EIO_READAHEAD;
1350	req->sv1 = newSVsv (fh);
1351	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1352	req->offs = SvVAL64 (offset);
1353	req->size = length;
1354
1355	EIO_SEND;
1356	}
1357
1358	void
1359	aio_stat (SV8 *fh_or_path, SV *callback=&PL_sv_undef)
1360	ALIAS:
1361	aio_stat = EIO_STAT
1362	aio_lstat = EIO_LSTAT
1363	PPCODE:
1364	{
1365	dREQ;
1366
1367	req->ptr2 = malloc (sizeof (EIO_STRUCT_STAT));
1368	if (!req->ptr2)
1369	{
1370	req_destroy (req);
1371	croak ("out of memory during aio_stat statdata allocation");
1372	}
1373
1374	req->flags |= FLAG_PTR2_FREE;
1375	req->sv1 = newSVsv (fh_or_path);
1376
1377	if (SvPOK (fh_or_path))
1378	{
1379	req->type = ix;
1380	req->ptr1 = SvPVbyte_nolen (req->sv1);
1381	}
1382	else
1383	{
1384	req->type = EIO_FSTAT;
1385	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1386	}
1387
1388	EIO_SEND;
1389	}
1390
1391	void
1392	aio_utime (SV8 *fh_or_path, SV *atime, SV *mtime, SV *callback=&PL_sv_undef)
1393	PPCODE:
1394	{
1395	dREQ;
1396
1397	req->nv1 = SvOK (atime) ? SvNV (atime) : -1.;
1398	req->nv2 = SvOK (mtime) ? SvNV (mtime) : -1.;
1399	req->sv1 = newSVsv (fh_or_path);
1400
1401	if (SvPOK (fh_or_path))
1402	{
1403	req->type = EIO_UTIME;
1404	req->ptr1 = SvPVbyte_nolen (req->sv1);
1405	}
1406	else
1407	{
1408	req->type = EIO_FUTIME;
1409	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1410	}
1411
1412	EIO_SEND;
1413	}
1414
1415	void
1416	aio_truncate (SV8 *fh_or_path, SV *offset, SV *callback=&PL_sv_undef)
1417	PPCODE:
1418	{
1419	dREQ;
1420
1421	req->sv1 = newSVsv (fh_or_path);
1422	req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1423
1424	if (SvPOK (fh_or_path))
1425	{
1426	req->type = EIO_TRUNCATE;
1427	req->ptr1 = SvPVbyte_nolen (req->sv1);
1428	}
1429	else
1430	{
1431	req->type = EIO_FTRUNCATE;
1432	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1433	}
1434
1435	EIO_SEND;
1436	}
1437
1438	void
1439	aio_chmod (SV8 *fh_or_path, int mode, SV *callback=&PL_sv_undef)
1440	ALIAS:
1441	aio_chmod = EIO_CHMOD
1442	aio_fchmod = EIO_FCHMOD
1443	aio_mkdir = EIO_MKDIR
1444	PPCODE:
1445	{
1446	dREQ;
1447
1448	req->type = type;
1449	req->int2 = mode;
1450	req->sv1 = newSVsv (fh_or_path);
1451
1452	if (ix == EIO_FCHMOD)
1453	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1454	else
1455	req->ptr1 = SvPVbyte_nolen (req->sv1);
1456
1457	EIO_SEND;
1458	}
1459
1460	void
1461	aio_chown (SV8 *fh_or_path, SV *uid, SV *gid, SV *callback=&PL_sv_undef)
1462	PPCODE:
1463	{
1464	dREQ;
1465
1466	req->int2 = SvOK (uid) ? SvIV (uid) : -1;
1467	req->int3 = SvOK (gid) ? SvIV (gid) : -1;
1468	req->sv1 = newSVsv (fh_or_path);
1469
1470	if (SvPOK (fh_or_path))
1471	{
1472	req->type = EIO_CHOWN;
1473	req->ptr1 = SvPVbyte_nolen (req->sv1);
1474	}
1475	else
1476	{
1477	req->type = EIO_FCHOWN;
1478	req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1479	}
1480
1481	EIO_SEND;
1482	}
1483
1484	void
1485	aio_unlink (SV8 *pathname, SV *callback=&PL_sv_undef)
1486	ALIAS:
1487	aio_unlink = EIO_UNLINK
1488	aio_rmdir = EIO_RMDIR
1489	aio_readdir = EIO_READDIR
1490	PPCODE:
1491	{
1492	dREQ;
1493
1494	req->type = ix;
1495	req->sv1 = newSVsv (pathname);
1496	req->ptr1 = SvPVbyte_nolen (req->sv1);
1497
1498	EIO_SEND;
1499	}
1500
1501	void
1502	aio_link (SV8 *oldpath, SV8 *newpath, SV *callback=&PL_sv_undef)
1503	ALIAS:
1504	aio_link = EIO_LINK
1505	aio_symlink = EIO_SYMLINK
1506	aio_rename = EIO_RENAME
1507	PPCODE:
1508	{
1509	dREQ;
1510
1511	req->type = ix;
1512	req->sv1 = newSVsv (oldpath);
1513	req->ptr1 = SvPVbyte_nolen (req->sv1);
1514	req->sv2 = newSVsv (newpath);
1515	req->ptr2 = SvPVbyte_nolen (req->sv2);
1516
1517	EIO_SEND;
1518	}
1519
1520	void
1521	aio_mknod (SV8 *pathname, int mode, UV dev, SV *callback=&PL_sv_undef)
1522	PPCODE:
1523	{
1524	dREQ;
1525
1526	req->type = EIO_MKNOD;
1527	req->sv1 = newSVsv (pathname);
1528	req->ptr1 = SvPVbyte_nolen (req->sv1);
1529	req->int2 = (mode_t)mode;
1530	req->offs = dev;
1531
1532	EIO_SEND;
1533	}
1534
1535	void
1536	aio_busy (double delay, SV *callback=&PL_sv_undef)
1537	PPCODE:
1538	{
1539	dREQ;
1540
1541	req->type = EIO_BUSY;
1542	req->nv1 = delay < 0. ? 0. : delay;
1543
1544	EIO_SEND;
1545	}
1546
1547	void
1548	aio_group (SV *callback=&PL_sv_undef)
1549	PROTOTYPE: ;$
1550	PPCODE:
1551	{
1552	dREQ;
1553
1554	req->type = EIO_GROUP;
1555
1556	req_send (req);
1557	XPUSHs (req_sv (req, AIO_GRP_KLASS));
1558	}
1559
1560	void
1561	aio_nop (SV *callback=&PL_sv_undef)
1562	ALIAS:
1563	aio_nop = EIO_NOP
1564	aio_sync = EIO_SYNC
1565	PPCODE:
1566	{
1567	dREQ;
1568
1569	#endif
1570
1571	void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)	2721	eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1572	{	2722	{
1573	grp->int2 = limit;	2723	grp->int2 = limit;
1574	grp->feed = feed;	2724	grp->feed = feed;
1575		2725
1576	grp_try_feed (grp);	2726	grp_try_feed (grp);
1577	}	2727	}
1578		2728
		2729	void
1579	void eio_grp_limit (eio_req *grp, int limit)	2730	eio_grp_limit (eio_req *grp, int limit)
1580	{	2731	{
1581	grp->int2 = limit;	2732	grp->int2 = limit;
1582		2733
1583	grp_try_feed (grp);	2734	grp_try_feed (grp);
1584	}	2735	}
1585		2736
		2737	void
1586	void eio_grp_add (eio_req *grp, eio_req *req)	2738	eio_grp_add (eio_req *grp, eio_req *req)
1587	{	2739	{
1588	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	2740	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
		2741
		2742	grp->flags |= EIO_FLAG_GROUPADD;
1589		2743
1590	++grp->size;	2744	++grp->size;
1591	req->grp = grp;	2745	req->grp = grp;
1592		2746
1593	req->grp_prev = 0;	2747	req->grp_prev = 0;
…		…
1597	grp->grp_first->grp_prev = req;	2751	grp->grp_first->grp_prev = req;
1598		2752
1599	grp->grp_first = req;	2753	grp->grp_first = req;
1600	}	2754	}
1601		2755
		2756	/*****************************************************************************/
		2757	/* misc garbage */
1602		2758
		2759	eio_ssize_t
		2760	eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
		2761	{
		2762	return eio__sendfile (ofd, ifd, offset, count);
		2763	}
		2764

Diff Legend

–	Removed lines
+	Added lines
<	Changed lines
>	Changed lines